The Guilty Look in Dogs: What the Behavior Actually Signals

1. What Is Being Claimed

1.1 The Everyday Reading

Owners report the guilty look consistently and describe it in similar terms: lowered body and head, ears back, averted gaze, tail tucked or wagging low, sometimes rolling onto the back, sometimes leaving the room. The interpretation attached to it is that the dog evaluated its own action against a known rule and found itself in breach.

That interpretation is not trivial. It requires the dog to hold a rule in memory, compare a past action to it, and produce an emotional response to the mismatch.

1.2 What Guilt Requires

Guilt is classed as a secondary or self-conscious emotion, meaning it depends on some representation of the self and of a standard the self has violated. Primary emotions — fear, anger, joy — are widely accepted across mammals (with the underlying neurochemistry reasonably well described). Secondary emotions are contested, and the evidence for them outside primates is thin.

When 907 animal owners were surveyed about emotions they had observed, primary emotions were reported more frequently than secondary ones, though jealousy was reported at strikingly high rates: 81 percent for dogs and 79 percent for horses (Morris, Doe & Godsell, 2008). That figure demonstrates how readily owners attribute complex emotions — not that the emotions were present.

1.3 Why This Matters Practically

If the display signals guilt, then scolding is informative to the dog. If it signals something else, scolding is at best noise and at worst a source of fear attached to the owner's return (with punishment carrying documented costs).

2. The Horowitz Experiment

2.1 The Design

Fourteen dogs were videotaped across a series of trials. In each, the owner showed the dog a treat, forbade it, and left the room. The experimenter then either removed the treat or allowed the dog to eat it. On the owner's return, the owner was told — accurately or inaccurately — whether the dog had obeyed, and greeted or scolded the dog accordingly (Horowitz, 2009).

The design separates two variables that are normally confounded: what the dog actually did, and what the owner believed and did in response. That separation is what makes the study informative, and it is the kind of operational discipline this whole area depends on (as measurement design generally does).

2.2 The Result

Behaviors associated with the guilty look showed no difference between dogs that had eaten the treat and dogs that had not. They appeared markedly more often in trials where the owner scolded — regardless of what the dog had done (Horowitz, 2009).

2.3 The Detail That Is Usually Omitted

The effect of scolding was more pronounced in the dogs that had been obedient. The dogs that had not eaten the treat and were scolded anyway produced the most intense displays (Horowitz, 2009).

That is the opposite of what a simple guilt account predicts, and it is one of the most informative findings in the study. If the display tracked wrongdoing, the innocent scolded dogs should have shown it least. They showed it most — consistent with a response to an unexpected and unexplained threat (which is how fear learning would predict it).

3. The Replication

3.1 A Larger Sample, a Different Setup

Sixty-four dogs were given the opportunity to break a rule — food on a table is for humans — while the owner was out of the room. On the owner's return, dogs that had eaten were no more likely to display guilty-look behaviors than dogs that had abstained. In this setup, the display was essentially absent without a scolding owner (Hecht, Miklósi & Gácsi, 2012).

3.2 Owners Cannot Tell

The same study tested whether owners could identify, from their dog's greeting behavior, whether it had broken the rule in their absence. They could not. Owners who had previously seen their dogs comply were no better at detecting transgression than anyone else (Hecht et al., 2012).

This is worth sitting with, because it contradicts the most common owner claim in this area — that they can tell immediately, before finding any evidence. Humans are demonstrably unreliable readers of canine signals more broadly (a pattern that recurs in reactivity cases).

3.3 What the Questionnaire Found

Alongside the experiment, a questionnaire established that the majority of owners perceive dog behavior as guilty in certain situations and believe dogs know when they have committed a disapproved act (Hecht et al., 2012). The belief is near-universal; the behavioral evidence for it is absent.

4. Without Scolding

4.1 The Remaining Objection

The obvious counter to both studies is that owners see the display before they scold, sometimes before they have noticed anything at all. If the display can appear without scolding, something else must trigger it — perhaps the dog's own action, perhaps the visible evidence of the misdeed.

4.2 The Test

Both variables were manipulated systematically. The experimenter either removed the forbidden food or let the dog eat it, and on the owner's return the food was either absent or visibly replaced. Owners were instructed to behave neutrally, and then asked to judge from their dog's greeting whether it had eaten (Ostojić, Tkalčić & Clayton, 2015).

4.3 The Result

Owners judged correctly no more often than chance. The dogs' greeting behavior was not affected by their own action, nor by the presence or absence of the food. The findings did not support the hypothesis that dogs show the guilty look in the absence of a concurrent negative reaction from the owner (Ostojić et al., 2015).

Three studies, three designs, one direction.

5. What the Display Actually Is

5.1 Conflict Reduction, Not Confession

The behaviors that make up the guilty look — lowered posture, averted gaze, flattened ears, tail tucked, lip-licking, rolling over — overlap strongly with recognised appeasement and conflict-reducing signals. The same set also appears under fear, uncertainty and conflicting motivation, which is part of the point: none of these behaviors is diagnostic of a specific internal state on its own.

Read that way, the display is not about the past at all. It is about the present: the owner is looming, tense, loud or staring, and the dog is doing what dogs do in that situation — a response shaped by its own history with that person (and by early social experience) (which is why reading emotion off a single posture fails).

5.2 The Cues Are Subtler Than Owners Realise

Owners who insist their dog displayed before any scolding are usually reporting accurately. What they are missing is their own behavior in the seconds before — the pause at the door, the sharp inhale, the change in gait on seeing the mess. Dogs read human posture, gaze and tone with precision (as the work on human gesture reading shows), and they are sensitive to human emotional state through channels the person is not monitoring (including odour).

5.3 Association, Not Deduction

A dog that has been scolded on returning to a chewed shoe several times does not need to reason about culpability. Shoe-plus-owner-arriving has become a predictor of an unpleasant event, and the appeasement display is a conditioned response to that predictor (by ordinary associative learning).

This also explains the case owners find most convincing: the dog that slinks off before the owner has seen anything. The mess is the cue. The dog did not need to have caused it.

6. Why the Display Persists

6.1 The Functional Answer

If the guilty look tells the owner nothing true, why does it happen at all? The questionnaire data in the replication study supplies an answer that is easy to miss: owners reported that when a dog displayed guilty behavior, they scolded it less (Hecht et al., 2012).

6.2 What That Means

The display may persist because it changes what the owner does next. That is a reinforcement account, and reinforcement requires no strategy on the dog's part — a behavior that is reliably followed by a milder outcome becomes more likely, whether or not the animal represents the connection. It is a sufficient explanation for the behavior's persistence without any appeal to guilt (the same logic that maintains other socially effective displays).

6.3 The Loop This Creates

Owner returns and reacts; dog appeases; owner softens. Both parties are reinforced, and the owner's belief that the dog "knew" is reinforced alongside. The interpretation survives because the interaction is genuinely working — just not for the reason assumed (a pattern that recurs wherever behavior is read as emotion).

7. What Would Be Needed to Show Guilt

7.1 The Bar

Demonstrating guilt would require showing that a dog represents a rule, represents its own past action, compares the two, and produces an affective response to the mismatch. No canine paradigm currently tests that chain, and each link is hard to isolate.

Dogs do show related capacities in fragments — they have episodic-like memory for their own recent actions, and they track human attentional states — but assembling those fragments into self-evaluation is a much larger claim (as the metacognition literature illustrates).

7.2 The Line These Studies Do Not Cross

This is where popular coverage overreaches. Headlines announcing that dogs feel no shame or no guilt state something the research did not test. What was tested is narrower: whether a specific display is evidence of guilt. It is not.

Whether dogs have any capacity for self-conscious emotion remains open, and absence of evidence in one paradigm is not evidence of absence (a distinction that matters throughout emotion research in dogs).

7.3 What Follows for Practice

The practical conclusion does not depend on resolving that question. Scolding a dog on returning to a mess does not inform it about the mess; it conditions an association with the owner's arrival. If the behavior is to change, it changes through management and training addressed to the moment the behavior occurs (with the emotional side handled separately), and destructive behavior in the owner's absence frequently has a different driver entirely (separation distress being the obvious candidate).

8. Summary at a Glance

The display tracks the owner, not the deed — Guilty-look behaviors showed no difference between dogs that had eaten the forbidden treat and those that had not, but appeared far more when owners scolded (Horowitz, 2009).

Innocent dogs displayed most — The effect of scolding was strongest in dogs that had actually obeyed, which is the opposite of what a guilt account predicts (Horowitz, 2009).

Replicated with 64 dogs — Dogs that broke the rule were no more likely to display than dogs that did not, and the display was essentially absent without a scolding owner (Hecht et al., 2012).

Owners cannot detect transgression — Owners were no better than chance at judging from greeting behavior whether their dog had broken a rule in their absence (Hecht et al., 2012; Ostojić et al., 2015).

Neither the act nor the evidence triggers it — With owners behaving neutrally, greeting behavior was unaffected by whether the dog had eaten or whether the food was visible (Ostojić et al., 2015).

The behavior set is conflict-reducing — It overlaps strongly with recognised appeasement signals, which also appear under fear and uncertainty; none of it is diagnostic of a specific internal state.

It may persist because it changes the owner's response — Owners reported scolding less when the dog displayed guilty behavior (Hecht et al., 2012), which is a sufficient reinforcement account and requires no strategy on the dog's part.

"Dogs feel no guilt" overstates it — The studies tested whether one display is evidence of guilt. They did not test whether dogs have any capacity for self-conscious emotion.

9. Research Gaps and Critical Appraisal

The samples are small. Fourteen dogs in the original study, sixty-four in the replication. The convergence across three independent designs is more persuasive than any one of them, but these are not large studies.

The paradigm is narrow. All three used forbidden food and brief owner absence. Whether the same pattern holds for other transgressions, longer absences, or contexts with different emotional weight has not been tested.

Absence of evidence is not evidence of absence. No study has tested whether dogs possess self-conscious emotion. What has been tested is whether one particular display indexes it, and the answer is no — which leaves the larger question untouched.

Conflict reduction is an interpretation, not a measurement. Reading the display as appeasement is well founded in ethological description and consistent with the data, but the studies measured behavior frequency, not function — and the same behaviors occur under fear and uncertainty without any social partner to appease.

Owner attribution data is self-report. The finding that owners scold less when dogs display is questionnaire-based (Hecht et al., 2012). It is a plausible reinforcement mechanism and has not been demonstrated experimentally.

Secondary emotion claims rest on owner report. The high attribution rates for jealousy and related states come from surveys of what owners believe they observed (Morris et al., 2008), which documents attribution rather than emotion.

There is no validated indicator of internal state. This limitation runs under everything here: no measure exists that would confirm or exclude a self-conscious emotion in a dog, which is why the question is likely to remain open (a constraint the judgement bias literature runs into as well).

10. Conclusion

Three experiments with different designs converge on the same finding, and it is a narrow one worth stating precisely: the guilty look is a response to what the owner is doing, not a report on what the dog did. The most telling result is not that guilty dogs and innocent dogs looked the same — it is that the innocent dogs scolded anyway produced the strongest displays of all, which is difficult to reconcile with any simple explanation based on the dog detecting its own wrongdoing. What the behavior more plausibly is, is conflict reduction: a set of signals overlapping heavily with appeasement, directed at a social partner who is currently signalling threat, and it may persist because it changes what that partner does next — owners themselves report scolding less when they see it. Two things follow. The practical one is that scolding on arrival teaches a dog about arrivals, not about shoes — and the behavior in question, if it happens in the owner's absence, needs to be addressed where and when it occurs. The second is a matter of intellectual honesty, and it cuts against the way this research is usually reported: none of these studies asked whether dogs can feel guilt. They asked whether one familiar display is evidence of it. Concluding from a negative answer that dogs have no inner life of that kind is a larger claim than anyone has earned (and the limits of behavioral inference are well documented).

Key Insights (Takeaways)

• The display responds to the owner's behavior, not the dog's. Across fourteen dogs, guilty-look behaviors did not differ between those that had eaten a forbidden treat and those that had not, but appeared markedly more often when owners scolded (Horowitz, 2009). Two independent studies with different designs reproduced the pattern (Hecht et al., 2012; Ostojić et al., 2015).

• Innocent dogs that were scolded showed the strongest displays. This is the result that decides the question: if the look indexed wrongdoing, obedient dogs should show it least, and they showed it most (Horowitz, 2009). It fits a response to unexpected threat and sits awkwardly with any account built on the dog evaluating its own past action.

• Owners cannot tell, even though they believe they can. Owners judged no better than chance whether their dog had broken a rule in their absence, and owners who had watched their dogs comply were no better than anyone else (Hecht et al., 2012; Ostojić et al., 2015). The near-universal claim of being able to tell immediately is not supported.

• The display may persist because it changes what the owner does next. Owners reported scolding less when their dog showed guilty behavior (Hecht et al., 2012). That is a reinforcement account rather than a strategy on the dog's part, and it explains the behavior's maintenance without invoking guilt — while also explaining why the owner's interpretation feels confirmed each time.

• "Dogs feel no guilt" is not what the research showed. These studies tested whether one display is evidence of guilt, and it is not. Whether dogs possess self-conscious emotions of any kind has not been tested, and no validated measure of internal state currently exists that could settle it.

References

Hecht, J., Miklósi, Á., & Gácsi, M. (2012). Behavioral assessment and owner perceptions of behaviors associated with guilt in dogs. Applied Animal Behaviour Science, 139(1–2), 134–142. https://doi.org/10.1016/j.applanim.2012.02.015

Horowitz, A. (2009). Disambiguating the "guilty look": Salient prompts to a familiar dog behaviour. Behavioural Processes, 81(3), 447–452. https://doi.org/10.1016/j.beproc.2009.03.014

Morris, P., Doe, C., & Godsell, E. (2008). Secondary emotions in non-primate species? Behavioural reports and subjective claims by animal owners. Cognition & Emotion, 22(1), 3–20. https://doi.org/10.1080/02699930701273716

Ostojić, L., Tkalčić, M., & Clayton, N. S. (2015). Are owners' reports of their dogs' "guilty look" influenced by the dogs' action and evidence of the misdeed? Behavioural Processes, 111, 97. https://doi.org/10.1016/j.beproc.2014.12.010