Researchers at the University of Parma’s dog ethology laboratory recruited 38 dogs for a study published in Royal Society Open Science. Each one was settled on a loose leash in a three by four metre room and played a forty second film of three people looking straight into the camera.
Two of the three faces were strangers. The third was the dog’s own owner, filmed at the laboratory after the dog’s five minute habituation, and then sitting on a chair beside the projector under instructions to ignore the dog. The owners had not been told what the experimenters were counting.
They were counting blinks.
The design
Each dog saw two versions of the film, in an order counterbalanced across the sample, with a five minute break between them. In the BLINK version each person blinked four times, once every three seconds, across their twelve seconds of screen time. In the CONTROL version the same person held a still, neutral gaze for the same twelve seconds.
The blinks were real. Rather than ask anyone to perform a blink, the experimenter recorded about two minutes of natural footage of each owner’s face, then cut individual spontaneous blinks out of it and stitched them into a rhythm. The unfamiliar actors were recorded and processed the same way. A green cone appeared with a beep at the start of the film, between each face and at the end, to pull the dog’s attention back to the screen.
Four unfamiliar actors were used in all, organised into two fixed pairs of one man and one woman, so that no result would rest on a single stranger’s face. Each dog met the same pair in both conditions, and the pair a dog was assigned made no significant difference to how it behaved. Twelve combinations covered every order of person and every order of condition.
Coding was done from two GoPro cameras, one in front of the dog and one to the side to catch the eyes of dark coated animals, using the Dog Facial Action Coding System. Two certified coders scored the footage, and a coder blind to the condition rescored a fifth of it, agreeing with them at correlations between 0.70 and 0.92.
The blink counts
Three dogs never met the inclusion bar. The bar was a simple one: they had to face the screen with their eyes open for at least four seconds in both conditions, long enough to have seen one blink. That left 35 for analysis, which is also the number the paper’s own abstract uses for the dogs it says were exposed to the videos.
Averaged across all three people, the dogs blinked at a higher rate watching the blinking faces than the still ones. The model that included the condition beat the null model, and the effect was significant.
Then the researchers split the result by who was on screen, and the picture changed. The gap between the blinking and still versions was significant when the face was the dog’s own owner, clearly for female owners (p = 0.004) and more narrowly for male owners (p = 0.046). For the two strangers there was no significant gap at all.
Compare across people within the blinking films, though, and only the female owners separate from the strangers. Dogs blinked more at them than at either stranger. Male owners did not differ significantly from either stranger, and female and male owners did not differ significantly from each other. Owner sex is also a between-dogs variable here: 20 of the 38 recruited had a female owner and 18 had a male one, which after the three exclusions left 19 and 16 in the analysis, and no dog saw both.
The paper’s abstract compresses this into one clause, reporting higher blink rates towards owners than unfamiliar humans, before adding the female-owner result in its next sentence. The figures above follow the results section, in which male owners did not separate from the strangers.
So the headline is not that dogs blink back at people in general. It is that the blink-video effect showed up for the dogs’ own people, and that the only person who separated from a stranger on a direct comparison was a female owner.
The prediction the study could not confirm
The third of the study’s three predictions was more specific than a rate. If blinking is a genuine piece of social coordination, the dogs should have timed their blinks to the humans’, and most tightly to their owner’s.
Not measurably. Across 96 dog blinks scored while the animals were facing the screen during the blinking films, 55 fell within a second and a half after a human blink and 41 did not. That is 57.3 per cent, which the authors themselves call a slight majority, and a one sided test put it at p equal to 0.092, not significantly greater than chance. The statistical model found no significant difference between owners and strangers either.
The authors report the limitation themselves. They ran a power analysis afterwards and put it at 0.38, meaning that even if a real synchrony effect of this size existed, this experiment would have missed it more often than it found it. That figure is computed from the observed 57.3 per cent itself, so by our reading rather than the authors’ it restates the non-significant result in different units rather than adding independent evidence about the experiment’s sensitivity.
Why anyone thought to look
Blinking already had form in the comparative literature. In humans, spontaneous blinks fall into step during conversation and while watching the same film, and blink rate rises with group size in several species, among them red deer, olive baboons and red junglefowl. In a study of four adult male rhesus macaques shown videos of other macaques, two of the four entrained their blinks to the animals on screen, the two that returned the videoed monkeys’ gaze. Cats that receive slow blinks from an unfamiliar person are more likely to approach that person afterwards, and in shelter cats slow blinking has been linked to quicker adoption.
For dogs specifically, the same Parma group reported in 2025 that dogs blink more at videos of other dogs blinking than at videos of dogs licking their noses, with only a non-significant trend in the same direction against dogs holding still. A separate 2026 study from the group found that dogs blink more with their owner than with a familiar trainer. In 2022 Koyasu and colleagues reported that dogs and cats do synchronise their blinks with humans, in live interaction.
That last result is why the new null matters.
The gap between the two studies
This study reads very differently against the 2022 live study than it does alone. Live mutual gaze produced blink synchrony in 2022; a screen produced no detectable synchrony in 2026.
The authors offer two readings of that gap, and they decline to choose between them. One is that pre recorded video cannot provide the real time feedback that synchronisation may need. The other is their own power figure: 96 blink events was not enough to detect an effect of this size, and they say plainly that more dogs or more recorded blinks would be needed. Those explanations have very different implications, and on this data the screen-versus-live account is not testable at all, which is a fact about the design rather than a finding about dogs. The authors do not treat the earlier result as settled either, since their own introduction notes that the live studies ran through nets or windows, which they say makes it impossible to exclude barrier effects or frustration.
Blinking in dogs has been linked to frustration as well as to affiliation, and a dog watching a screen while its owner pointedly ignores it is not obviously in a relaxed state. The owner gender effect is flagged by the authors themselves as exploratory, since the study was not designed to disentangle an owner’s sex from interaction history or caregiving routine. Among the other behaviours the coders scored, including nose licking, lip wiping and looking away, none showed a significant difference between conditions or between people. The authors read that as support for the idea that blinking is a highly sensitive response, while other facial cues may need greater emotional arousal or closer, more immersive contact before they appear. And every dog in the study was reacting to a recording rather than to a live person.
The narrow result
What survives is narrow and specific. Thirty-five dogs, watching a screen, blinked at a significantly higher rate when the person on it was blinking, and broken down by person that gap appeared for their own people and not detectably for the strangers.
Somewhere in Parma there is a dog on a loose leash, watching a green cone appear with a beep and then its owner’s face arriving for twelve seconds. Its owner is on a chair beside the projector, saying nothing. The dog blinks. What that blink means is still, on the evidence here, an open question.