Emotions in Dogs: What We Know and What We Do Not

1. A Question That Cannot Be Answered Directly

1.1 The Sentence to Start From

Kujala's review of canine emotion opens with a statement that sets the terms for everything that follows: it is not possible to demonstrate that dogs feel emotions, but the same is true for all other species, including our own.

That is not skepticism about dogs. It is a point about the kind of thing a feeling is. No experiment reaches another being's subjective experience, human or otherwise; what we have with other people is verbal report, which is testimony rather than observation. Remove the report and the problem with dogs is the problem with a pre-verbal infant, and it has to be approached the same way — indirectly, from what can be observed, with premises borrowed from the human case.

The consequence is that "do dogs have emotions?" is not a question research can settle by finding better equipment. The productive questions are narrower: what internal states can be shown to exist, what can be measured about them, which claims the evidence supports, and where the human doing the interpreting is adding something that is not there.

1.2 What This Article Covers

This article sets out the two frameworks the field uses to carve up emotion and why they are not rivals, the distinction between an emotion state and a conscious feeling that makes the whole area tractable, what can and cannot currently be measured in a dog, where popular claims about jealousy and guilt outrun the data, how human social cognition distorts the interpretation in both directions, and what any of this changes in practice.

It is the foundational article beneath a number of more specific ones: fear, frustration, arousal, contagion and the measurement problem each have their own treatment and are linked where they belong.

2. Two Ways to Carve Up Emotion

2.1 Discrete Systems

One tradition treats emotions as a small set of distinct systems, each with its own neural circuitry and its own characteristic behavior. Panksepp's version is the most influential in animal work. He argues that primal affects are intrinsic brain value systems that inform an animal, automatically and without learning, how it is faring with respect to survival — positive affects indexing conditions that support survival, negative affects signaling conditions that may impair it — and that they are essential to emotional learning.

He names seven primary-process systems, deliberately capitalized to mark them as neural systems rather than as the English words: SEEKING, RAGE, FEAR, LUST, CARE, PANIC/GRIEF, and PLAY. In his account these arise from ancient caudal and medial subcortical regions rather than from cortex.

The practical appeal for anyone working with dogs is obvious. SEEKING covers the forward-leaning, investigating, anticipating state that dominates a working dog's day and is not the same as pleasure at obtaining something. PANIC/GRIEF is separation distress as a system in its own right rather than a variant of fear (separation-related behavior). RAGE and FEAR are distinct, which matters because they produce different behavior toward the same trigger (aggression in dogs). PLAY is its own system rather than practice for something else (the neurobiology of play).

The limitation is equally clear: the seven are a theoretical proposal about neural organization, largely developed in other species, and the mapping from any one of them onto a dog's behavior in a given moment is an inference rather than a reading.

2.2 Dimensions

The other tradition describes emotional states not as categories but as positions in a space. Mendl, Burman, and Paul's framework uses two axes: valence, running from negative to positive, and arousal, running from low to high. Every state sits somewhere in that space, and the same level of arousal can accompany opposite valences.

This has two advantages for animal work. It requires no commitment about which discrete emotions a species has, which means it can be applied without settling an argument that cannot be settled. And it maps directly onto what the measurement tools actually deliver, because most physiological indicators track arousal.

The quadrants are also immediately useful. High arousal with negative valence is the panicking dog; high arousal with positive valence is the dog about to be thrown a ball; low arousal with negative valence is the withdrawn, shut-down dog; low arousal with positive valence is contentment. Two of those four are routinely confused with each other in practice, and the confusion runs in both directions (arousal regulation).

2.3 Why These Are Not Competitors

It is tempting to treat this as a dispute to be resolved. It is better treated as two descriptions at different grain.

The dimensional frame is the right tool for measurement, because it asks only for two quantities and both have candidate indicators. The discrete frame is the right tool for asking what a state is for — what behavior it organizes and what it is trying to achieve. A dog can be placed in the high-arousal negative quadrant by physiology; whether the state is organizing escape or defense is a question the quadrant does not answer and the systems view does.

In casework both are needed, and mixing them up produces a characteristic error: a measured arousal figure gets reported as if it identified an emotion.

3. The Distinction Everything Rests On

3.1 State Versus Feeling

Anderson and Adolphs proposed the conceptual move that makes emotion research in non-verbal animals possible. They separate the emotion state — an internal central nervous system state that causes observable behavior and physiological responses — from the subjective feeling. In humans the two occur together: central emotion states produce subjective feelings in parallel with behavioral and somatic responses. In a non-verbal organism the state can be investigated without the feeling being settled.

This is not a dodge. It relocates the question. Whether a dog's fear state is accompanied by something it is like to be that dog remains open; that the dog has a fear state, with properties that can be tested, does not.

3.2 Four Properties That Make a State an Emotion

Anderson and Adolphs also specify what would qualify an internal state as emotional rather than reflexive. They propose four emotion primitives.

Scalability. The state varies in intensity, from mild to intense — annoyance through to rage — and the behavior it produces can change qualitatively as intensity rises.

Valence. States come in opposing pairs expressed through physically contrasting behavior, following Darwin's principle of antithesis: approach against withdrawal, aggression against courtship.

Persistence. The state outlasts the stimulus that triggered it and continues to influence behavior and cognition afterwards. This is the property that most clearly separates an emotion from a reflex.

Generalization. A state induced by one stimulus carries over into other contexts and biases later responses.

3.3 Why This Is the Most Useful Part of the Theory for Practice

Those four properties are not abstractions. They describe things anyone working with dogs sees constantly, and they explain why certain training assumptions fail.

Persistence is why a dog does not reset when the trigger leaves. The state is still running, and a request made two minutes later is being made to a different animal than the one from ten minutes before.

Generalization is why a bad encounter in one place shows up somewhere unrelated later the same day, and it is the mechanism underneath the stacking of several manageable events into one unmanageable state (trigger stacking).

Scalability is why the same trigger produces orienting on one day and a full display on another, without anything having changed about the trigger.

And valence as antithesis is why two dogs producing similar-looking behavior on a lead may be in opposite states — which is the everyday version of the point that behavior does not read off as emotion (why behavior is not emotion).

3.4 The Componential View

Paul, Sher, Tamietto, Winkielman, and Mendl add the structural point. The componential view of human emotion recognizes that affective states comprise conscious, behavioral, physiological, neural, and cognitive elements. Their question is whether all those components operate similarly across taxonomic groups, and they argue for developing scientific methods to identify markers of conscious emotional processing rather than treating the question as unanswerable in principle.

Two things follow. First, an emotional state is not one thing that a single measure could capture; it is several components that can dissociate. Second, the consciousness question is being treated as a research program rather than as a permanent mystery — which is a more optimistic position than the one this article started from, and it is a live disagreement in the field rather than a settled point.

3.5 Emotion and Mood Are Different Timescales

Mendl, Burman, and Paul's framework is titled as being about animal emotion and mood, and the second term does work that is easy to skip.

An emotion in their sense is tied to an event: it arises in response to something, has a target, and is relatively short-lived. A mood is a longer-running background state, not tied to a particular stimulus, that biases how the next event is appraised. The two are related — moods are built in part from accumulated emotional episodes — but they are not the same object, and they behave differently.

The distinction has three practical consequences.

It explains the inconsistency owners report most often. A dog that handles a passing dog calmly on Tuesday and reacts on Thursday has not forgotten its training. The event is the same; the background state it lands in is not. Persistence and generalization operate at the level of the episode, mood operates above it, and the two compound.

It is what judgment-bias testing actually measures. The optimism-pessimism paradigm presents an ambiguous stimulus and records which way the animal resolves it. That resolution reflects the background expectation the animal brings rather than an emotion about the stimulus itself, which is why judgment bias is among the more interesting measures for welfare purposes and among the least useful for answering what a dog is feeling at a given moment.

It changes what an intervention aims at. Reducing the intensity of single episodes and shifting the baseline a dog operates from are different goals, with different methods and different timescales. For a dog living with chronic fear the second is usually where the difference is made, and it is measured in weeks rather than sessions (chronic stress).

Keeping the two apart also disciplines the language. "He is anxious" and "he was anxious about the bin lorry" are claims at different levels, and a plan built for one will not work if the other is the problem.

4. What Can Actually Be Measured

4.1 Physiology Reads Arousal, Mostly Under Threat

Flint, Weller, Parry-Howells, Ellerby, McKay, and King ran the comparison that the dimensional framework demands. Sixty healthy adult dogs across three breeds went through six ten-minute scenarios designed to produce four combinations of valence and arousal: toy play and treat throwing for positive high arousal, petting and a long-lasting chew for positive low arousal, car travel for negative high arousal, and social separation for negative low arousal. Measures included cortisol, ACTH, heart rate, heart-rate variability, eye, nose and ear temperature, serotonin, secretory immunoglobulin A, panting, whining, body shaking, posture, activity, and qualitative behavioral assessment.

Cortisol, ACTH, heart-rate variability, panting, whining, and body shaking showed arousal-related differences primarily within the negative conditions. Heart rate, secretory IgA, standing and lying varied with arousal across both valences. Qualitative behavioral assessment — a structured human judgment — was the best indicator of positive valence. The authors recommend combining parameters covering both valence and arousal, and note that food provision, handler expectations, exercise, and ambient temperature all influenced results.

The bottom line for anyone reading a cortisol figure: it tells you something happened, not whether the dog wanted it to (measuring stress in dogs).

4.2 Behavior: Used Constantly, Validated Rarely

De Winkel and colleagues systematically reviewed the observational behaviors and emotions used to assess dog welfare. They found 39 studies and grouped the indicators into nine themes. Five studies reported any form of validity assessment, 23 reported inter-rater reliability, and only six explicitly linked the observed behaviors to a welfare conclusion.

Behavioral indicators are the backbone of applied work with dogs and the least validated part of it. That is not an argument against using them — there is nothing better — but it is an argument for describing what was seen rather than what it meant (reading canine body language).

4.3 Facial Expression: Promising, and Routinely Overclaimed

The most methodologically careful attempt to read canine emotion from the outside uses the Dog Facial Action Coding System, which codes individual muscle movements rather than global impressions.

Bremhorst, Mills, Würbel, and Riemer tested 28 dogs, mostly Labrador Retrievers, across two reward types — food and toys — comparing positive anticipation against frustration. Movement of the ear adductor was associated with positive anticipation. Ear flattening, blinking, lip parting, jaw dropping, and nose licking appeared during frustration. Most of these expressions were context-independent, holding across both reward types, which is a genuine finding: the same facial actions accompanied the same state whether the reward was food or a toy (the neurobiology of frustration).

Then comes the sentence that should accompany every popular article about reading dogs' faces. Evaluated as diagnostic indicators, the authors concluded that none of the individual expressions would allow consistent correct classification of emotional state.

So: real, reproducible differences in facial movement between two emotional conditions, and no single facial movement that tells you which condition a dog is in. Those two findings are not in tension; they are what a weak but non-zero signal looks like. A nose lick is more common in frustration and also happens in plenty of other states, which is exactly why it cannot be used on its own (behavior is not emotion).

4.4 Cognitive Measures

One family of measures approaches valence differently: instead of reading the body, it asks the animal to make a judgment. In judgment-bias testing an ambiguous stimulus is presented and the animal's optimistic or pessimistic treatment of it is recorded, on the reasoning that mood biases expectation (cognitive bias in dogs).

It is slow, requires training, and is not a field method. It is also the approach that targets most directly the thing the physiological measures cannot reach, and it belongs in any list of what the field can currently do.

4.5 The Problem Under All of It

Cobb, Jiménez, and Dreschel identify construct validity as the central unsolved issue on the physiological side: measures are routinely treated as indicators of welfare or emotion without a demonstration that they index the construct claimed. The same charge applies to behavioral indicators, as De Winkel's validity count shows.

Construct validity is the unglamorous heart of this entire field. It is not about whether cortisol can be measured accurately — it can — but about whether the thing it measures is the thing being talked about.

5. Where the Claims Outrun the Evidence

5.1 Jealousy

Jealousy is the best-studied of the so-called secondary emotions in dogs, and the study designs are careful enough to show exactly where the interpretation stops.

Abdai, Baño Terencio, Pérez Fraga, and Miklósi tested 24 dogs from multi-dog households across five trials with four kinds of test partner: a familiar dog from the same household, an unfamiliar dog, a familiar object (a newspaper), and an unfamiliar object (a remote-control car or a thermos). They measured looking duration, body positioning, proximity to the owner, and attempts to interrupt during ninety-second test phases.

The dogs showed significantly more owner-directed behavior and more attempts to separate the owner from the social partners than from the non-social ones. The authors' framing is precise: this demonstrates jealous behavior, functioning similarly to what is observed in human infants under two years of age, while the underlying emotional experience remains uncertain.

That distinction is the whole point. The study establishes that the behavior is directed at social rivals specifically, not at anything that takes the owner's attention. It does not establish that the dog experiences what an adult human means by jealousy, which involves a representation of a relationship under threat from a third party and an appraisal of one's own standing in it (inequity aversion: fairness or jealousy?).

5.2 Guilt

The so-called guilty look is a clear example where the evidence redirects the interpretation rather than confirming the emotional label: the behaviors owners read as guilt track the owner's own behavior more closely than they track whether the dog did anything (the guilty look). Guilt as humans understand it requires a self-evaluation against a standard, and nothing in the canine literature reaches that.

5.3 The Inference That Keeps Being Made

The pattern across all the secondary-emotion claims is the same, and it has a name worth having: behavior that functions like X is not the same as feeling X.

A dog that pushes between its owner and another dog is doing something with a clear function — maintaining access to a valued social partner. Calling the state behind it jealousy imports a specific internal structure that the behavior does not require. The same applies to spite, guilt, shame, and embarrassment. In each case a simpler description covers the observation, and the richer one is added by the observer.

This is not a claim that dogs have only simple states. Anderson and Adolphs' four primitives already describe something considerably more than reflex, and Adolphs, in a commentary on exactly this question, takes the view that dogs do consciously experience emotions and that the interesting question is which ones. The point is narrower: a particular rich label is not licensed by a behavior that a leaner one also explains.

6. The Human in the Loop

6.1 We Attribute Minds Very Readily

Kujala makes the methodological warning that should sit alongside every finding in this area: humans are a socially minded species, and we readily impute mind and emotion to others — even to vegetables or rocks. Experimental results therefore need careful analysis, or the emotional lives of dogs get estimated wrongly.

The implication is not that every attribution is wrong. It is that our attributions are generated by a system that runs automatically and does not distinguish between a dog and a cartoon. Whatever it produces has to be checked rather than trusted.

6.2 What Owners Notice, and What They Do Not

Buckland, Volk, Burn, and Abeyesinghe surveyed 445 owners in the UK and Ireland about how they recognize positive emotional states in their dogs. Owners described two clusters. High-arousal positive states — "happy and excited" — were linked to anticipation of walks or food and described in terms of barking, raised head, wide bright eyes, pricked ears, tail wagging, open mouth, and active play. Low-arousal positive states — "happy and relaxed" — were linked to resting and gentle stroking and described as silence, head down, eyes closed or nearly closed, ears down, mouth closed.

The finding that matters: the low-arousal positive states were described in considerably less detail than the high-arousal ones, which the authors suggest indicates they are less well understood. Breed, age, owner experience, and level of attachment all influenced which indicators owners reported.

So the human attribution system is not uniformly generous. It is generous about excitement and thin about contentment — which means an assessment resting on owner impressions will systematically underweight the quiet positive states that make up most of a well-functioning dog's day (welfare science and quality of life).

6.3 The Error Runs Both Ways

Anthropomorphism is the error that gets named. The opposite error is real too and gets named less: denying a state because the human label for it would be embarrassing.

Kujala's own list of what has been shown is substantial. Dogs attend to social cues and respond appropriately to the valence of human and dog facial expressions and vocalizations of emotion. Their limbic reward regions respond to the odor of their caretakers. They behave differently according to the emotional situation, show emotionally driven expectations, have affective disorders, and exhibit some subcomponents of empathy. Their brain includes a relatively large prefrontal cortex and, as in primates, an area specialized for face perception. Her summary is that dogs have many degrees of emotion while the full extent remains unknown.

Treating that as nothing because it falls short of proving subjective experience is the mirror image of treating a nose lick as guilt. Both substitute a stance for an assessment (emotional contagion, vocal communication).

7. Where the Field Actually Stands

7.1 What Is Established

Dogs show internal states that satisfy several criteria used to identify emotional states rather than simple reflexes: they vary in intensity, show different valences, can persist beyond the triggering event, and can influence responses in other contexts. Those states can be moved reliably by experimental manipulation, they show up in physiology, posture, facial movement and cognitive bias, and they differ systematically between conditions designed to be positive and negative.

Affective disorders exist in dogs as clinical entities, respond to intervention, and have neurobiological correlates (the neurobiology of anxiety, psychopharmacology in dogs). That is not a small set of facts.

7.2 What Is Contested

Whether those states are consciously experienced is genuinely open, and the field does not speak with one voice. Kujala's position is that it cannot be demonstrated, for dogs or for anyone. Adolphs' is that dogs do consciously experience emotions and that the useful question is which. Paul and colleagues treat the identification of consciousness markers as a research program to be pursued rather than a question to be set aside.

Anyone writing confidently on either side of that is going beyond the literature.

7.3 What Is Not Established

Which discrete emotions a dog has, in the sense of a defensible list, is not established. Whether the secondary emotions requiring self-representation occur at all is not established. No single measure diagnoses an emotional state. And the construct validity of most indicators in use has not been demonstrated.

8. Emotion and Learning Are Not Separate Systems

8.1 Affect as the Signal Learning Runs On

Panksepp's claim that the primal affects serve an essential function in emotional learning is not a side remark. In his account the affective systems are what tell an animal whether what just happened was good or bad, and that signal is the input learning operates on. Remove it and there is nothing for reinforcement to be built from.

Read that way, emotion is not a layer sitting on top of a learning mechanism; it is part of the mechanism. Which means the common framing in training — that a dog either has a training problem or an emotional problem — is a rough triage rather than a real division (learned behavior versus emotional response).

8.2 Why the Triage Is Still Useful

The distinction survives as a practical one because the two cases call for different first moves.

Where the obstacle is that the dog does not know what is wanted, the work is instructional: clearer criteria, better timing, easier steps. Where the obstacle is that the dog is in a state incompatible with learning, instruction does not reach it, and the first job is changing the state.

The four primitives explain why. A state that persists past the stimulus and generalizes across contexts is not something a clearer cue fixes, and a state that scales means the same procedure can work at one intensity and fail at another. That is the structural reason why threshold matters in behavior work and why the same protocol run one step too fast produces no learning at all.

8.3 What This Implies About Aversive Methods

If affective states are the substrate learning runs on, then a method that works by generating a negative state is not simply supplying information. It is changing the animal's appraisal of the situation, and by the persistence and generalization properties, that change does not stay where it was put.

That is a mechanistic reason to expect effects beyond the trained behavior, and it is independent of any outcome study — though the outcome evidence points the same way (punishment fallout, the neurological picture).

8.4 Emotional Memory Is Not Ordinary Memory

Fear learning in particular has properties that ordinary instrumental learning does not: it forms fast, sometimes in a single trial, and extinguishing it suppresses the response without deleting the original association, which is why it returns (fear conditioning and reconsolidation, why extinguished behavior returns).

For practice this means a relapse after apparent success is the expected shape of the process rather than evidence that the work failed.

9. What Follows for Practice

9.1 Describe the Observation, Then the Inference

"She lip-licked twice, turned her head away, and stopped taking food for about ninety seconds" and "she was anxious" are different kinds of statement. Both can appear in a case record; only the first can be compared with next month's (operationalizing behavior, behavioral assessment).

9.2 Use Valence and Arousal as the Working Frame

For most practical purposes the two-axis description is enough and is less likely to be wrong. Asking whether the dog is highly aroused and whether the state looks positive or negative produces a usable picture without committing to a label the evidence would not support.

9.3 Never One Measure

This is the single most consistent recommendation across the literature, from Flint's experimental comparison to the reviews of behavioral and physiological indicators. One cortisol value, one posture, or one facial action alone does not identify a specific emotional state. Two or three independent lines, interpreted together and with the disagreements noted, is the standard.

9.4 Look for the Quiet Positive States Deliberately

Given that owners describe low-arousal positive states in far less detail than excited ones, contentment has to be looked for rather than waited for. Where does this dog settle, how long does it take to settle, does it sleep through, does it choose to be near people when nothing is happening? Those questions get at the states that an excitement-focused assessment misses.

9.5 Do Not Let the Label Become the Argument

Whether a particular dog "is jealous" or "is being possessive" or "is resource guarding the owner" is usually not a question about the dog. The behavior, the situations that produce it, and what changes it are the same whichever word is used, and the plan does not depend on winning the terminology (resource guarding).

9.6 What to Say to an Owner

Owners talk about their dogs in emotional terms and should not be corrected out of it — the vocabulary is how people care about animals, and an owner who says the dog is sad is usually reporting something real, if imprecisely. The useful move is not to deny the emotion word but to ask what the dog was doing when they noticed it. That converts an attribution into an observation without dismissing it, and it is usually where the information is.

10. Summary at a Glance

The question cannot be answered directly, for any species. Subjective feeling is not observable; with humans we accept verbal report instead. Remove report and dogs pose the problem a pre-verbal infant poses.

Separate the state from the feeling. An emotion state is an internal nervous-system state that causes behavior and physiological responses. Whether it is accompanied by subjective experience is a separate question, and only the first is currently testable in dogs.

Four properties separate an emotional state from a reflex. Scalability, valence, persistence, generalization. Dogs' states have been shown to vary in intensity and valence, to persist beyond the trigger, and to carry over into other contexts.

Two frameworks, two jobs. Seven discrete primary-process systems describe what a state is organizing; valence and arousal describe where it sits and map onto what can be measured.

Physiology mostly reads arousal, and mostly under threat. In sixty dogs across four valence-arousal combinations, cortisol, ACTH, heart-rate variability, panting, whining and body shaking differentiated arousal chiefly within negative conditions; structured human judgment was the best indicator of positive valence.

Behavioral indicators are the backbone and the least validated part. Of 39 studies reviewed, five reported any validity assessment and six linked the observed behaviors to a welfare conclusion.

Facial coding finds real differences and no diagnostic sign. In 28 dogs, specific facial actions distinguished positive anticipation from frustration and held across reward types — yet none would allow consistent correct classification on its own.

Jealous behavior is demonstrated; jealousy is not. Twenty-four dogs directed more owner-directed and separating behavior at social rivals than at objects, and the authors state that the underlying emotional experience remains uncertain.

The human attribution system is not neutral. We impute mind readily, and unevenly: owners described excited positive states richly and relaxed positive states vaguely.

The consciousness question is live, not closed. One position holds it cannot be demonstrated for any species; another holds that dogs do consciously experience emotions and the question is which; a third treats consciousness markers as a research program.

Emotion and mood are different timescales. An emotion is tied to an event; a mood is a background state that biases how the next event is appraised, and judgment-bias testing measures the second.

Affect is part of the learning mechanism, not a layer on it. The affective systems supply the good-or-bad signal reinforcement is built from, which is why a state incompatible with learning cannot be fixed with clearer instruction.

Construct validity is the field's central unsolved problem. Not whether the measures are accurate, but whether what they measure is what is being claimed.

11. Research Gaps

No validated indicator of positive valence. The leading candidate is a structured human judgment rather than an objective readout, and whether any physiological measure indexes positive states specifically is unresolved.

No defensible list of discrete canine emotions. The seven-system proposal was developed largely in other species. Which systems are separable in dogs, by dog data, has not been worked out.

Facial coding needs combinations, not single actions. Given that no individual action classifies reliably, the open question is whether configurations of several do — and that has not been tested at the scale it needs.

Secondary emotions lack a test that could come out negative. The jealousy studies show social specificity; no current design distinguishes an appraisal-based state from a simpler motivational one, which means the interesting claim is not being tested.

Consciousness markers are proposed rather than established. The program for identifying them across species exists as a proposal; no marker has been validated in dogs.

Almost everything is measured in short laboratory episodes. Emotional states that persist for hours or recur over weeks are the ones that matter for welfare, and the measurement literature is built on ten-minute scenarios (chronic stress).

12. Conclusion

The state of knowledge here is more definite than the popular treatment suggests in one direction and much less definite in the other.

More definite: dogs have internal states that behave like emotions by explicit criteria, that can be moved experimentally, that show up across physiology, posture, facial movement and cognitive bias, and that go wrong in ways recognizable as clinical disorders. Anyone who treats canine emotion as unscientific speculation is a couple of decades behind.

Much less definite: which emotions, how they are individuated, whether any of it is consciously experienced, and how to tell from the outside which state a particular dog is in right now. The most careful study of canine facial expression found genuine differences between two emotional conditions and concluded that no single expression classifies reliably. That result, rather than any headline, is a fair picture of the field.

What survives is a working discipline. Describe what the dog did. Place the state on two axes rather than naming it. Use more than one measure and say where they disagree. Look for the quiet positive states as hard as the loud negative ones. And hold the emotion word loosely, because the plan does not depend on it and the evidence does not license it.

Key Insights

You cannot prove a dog feels anything — and you cannot prove it of a person either. That symmetry is the starting point, not a reason to stop.

The state is testable; the feeling is not. Keeping those apart is what turned animal emotion from a philosophical argument into an experimental field.

Emotions persist and generalize. A dog does not reset when the trigger leaves, and a state from one context biases behavior in another. Most training failures blamed on stubbornness are one of these two.

Valence and arousal are enough for most purposes. Two axes, both with candidate indicators, no commitment to a label the evidence would not support.

Cortisol tells you something happened, not whether the dog wanted it to. Arousal measures do not distinguish thrilled from terrified.

There is no tell. The best facial-coding study found real differences between positive anticipation and frustration and no single action that classifies a dog's state reliably.

"Jealous behavior" is the accurate phrase. The behavior is directed specifically at social rivals; what the dog experiences is not established by that.

Contentment is harder to see than excitement. Owners describe relaxed positive states vaguely, so those states have to be looked for on purpose.

Anthropomorphism has an opposite, and it is also an error. Denying a documented state because its human label sounds sentimental is the same failure in reverse.

Tuesday calm and Thursday reactive is not forgotten training. The event was the same; the background state it landed in was not.

Ask what the dog was doing. The most useful response to "he was sad" is not correction but a question about the behavior behind it.

References

Abdai, J., Baño Terencio, C., Pérez Fraga, P., & Miklósi, Á. (2018). Investigating jealous behaviour in dogs. Scientific Reports, 8, 8911. https://doi.org/10.1038/s41598-018-27251-1

Adolphs, R. (2017). Dogs consciously experience emotions: The question is, which? Animal Sentience, 2(14), 2. https://doi.org/10.51291/2377-7478.1213

Anderson, D. J., & Adolphs, R. (2014). A framework for studying emotions across species. Cell, 157(1), 187–200. https://doi.org/10.1016/j.cell.2014.03.003

Bremhorst, A., Mills, D. S., Würbel, H., & Riemer, S. (2022). Evaluating the accuracy of facial expressions as emotion indicators across contexts in dogs. Animal Cognition, 25(1), 121–136. https://doi.org/10.1007/s10071-021-01532-1

Buckland, E. L., Volk, H. A., Burn, C. C., & Abeyesinghe, S. M. (2014). Owner perceptions of companion dog expressions of positive emotional states and the contexts in which they occur. Animal Welfare, 23(3), 287–296. https://doi.org/10.7120/09627286.23.3.287

Cobb, M. L., Jiménez, E. Y., & Dreschel, N. A. (2025). Beyond cortisol! Physiological indicators of welfare for dogs: Deficits, misunderstandings and opportunities. Journal of Applied Animal Welfare Science. Advance online publication. https://doi.org/10.1080/10888705.2025.2572616

De Winkel, T., van der Steen, S., Enders-Slegers, M.-J., Griffioen, R., Haverbeke, A., Groenewoud, D., & Hediger, K. (2024). Observational behaviors and emotions to assess welfare of dogs: A systematic review. Journal of Veterinary Behavior, 72, 1–17. https://doi.org/10.1016/j.jveb.2023.12.007

Flint, H. E., Weller, J. E., Parry-Howells, N., Ellerby, Z. W., McKay, S. L., & King, T. (2024). Evaluation of indicators of acute emotional states in dogs. Scientific Reports, 14, 6406. https://doi.org/10.1038/s41598-024-56859-9

Kujala, M. V. (2017). Canine emotions as seen through human social cognition. Animal Sentience, 2(14), 1. https://doi.org/10.51291/2377-7478.1114

Mendl, M., Burman, O. H. P., & Paul, E. S. (2010). An integrative and functional framework for the study of animal emotion and mood. Proceedings of the Royal Society B: Biological Sciences, 277(1696), 2895–2904. https://doi.org/10.1098/rspb.2010.0303

Panksepp, J. (2011). The basic emotional circuits of mammalian brains: Do animals have affective lives? Neuroscience and Biobehavioral Reviews, 35(8), 1791–1804. https://doi.org/10.1016/j.neubiorev.2011.08.003

Paul, E. S., Sher, S., Tamietto, M., Winkielman, P., & Mendl, M. T. (2020). Towards a comparative science of emotion: Affect and consciousness in humans and animals. Neuroscience and Biobehavioral Reviews, 108, 749–770. https://doi.org/10.1016/j.neubiorev.2019.11.014