Emotional communication facilitates social interactions among animals. While dogs can discriminate and exhibit context-appropriate behaviors in response to odor samples from stressed and non-stressed humans, their capacity to differentiate emotional odors from conspecifics—and the broader function of these cues in intraspecific communication—has not been investigated. Using a habituation-discrimination paradigm, 43 dogs were presented with odor samples from an unfamiliar dog collected after events marked as joyful, stressful, and baseline (relaxing). Subjects showed longer investigation time for novel versus repeated odors, indicating successful discrimination and dishabituation. Specifically, subjects distinguished between joy and baseline odors and between joy and stress odors. Additionally, dogs exhibited distinct behavioral responses depending on the emotional content of the odor, including differences in stress- and joy-related behaviors, proximity to targets (owner, stranger, and odor source), and body posture. Exposure to stress odors elicited the predicted attachment-related behaviors—characterized by increased proximity to their owners and reduced proximity to strangers—suggesting a response to conspecific emotional states that may reflect emotional contagion. These results demonstrate that dogs can perceive and behaviorally respond to emotional odors from conspecifics, underscoring the role of olfaction in canine intraspecific emotional communication.

Ethical approval

This research was approved by Columbia University IACUC and Barnard College IRB (AC-AABZ3660). Canine participation was voluntary, based on continued task engagement. Owners gave informed consent and could withdraw participation at any time.

Subjects

43 domestic dogs (23 M, 20 F) and their owners participated in at least one phase of the study. Inclusion criteria for subject dogs included having lived ≥6 months in their current home, having up-to-date vaccinations, being comfortable with novel environments, being treat-motivated, not having been discouraged from sniffing during walks, and having no olfactory-related medical issues. Mean subject age was 5.4 years (range: 10 mo to 13 yr 10 mo), all subjects were spayed/neutered except 2 males, 17 were purebred, 26 were mixed-breed (for specific breeds, see Part_1.xlsx), and each dog-owner pair participated only once.

Odor Stimuli

A 5-year-old spayed female domestic dog, unfamiliar to subjects, served as the odor donor for three emotional states: “joy,” “stress,” and “baseline.” Joy was elicited via 2 minutes of play involving ball retrieval; stress via 2 minutes of nail trimming while restrained; baseline via lightly resting alone. Saliva, interdigital, and perianal secretions were collected by the owner using cotton pads during a consistent 2-hour morning window. Samples were collected only when the dog was healthy and had not eaten for at least 2 hours. Cotton pads were cut into quarters, stored at −20°C in sterile labeled bags, and defrosted 30 minutes before use. Five minutes before the study, samples were warmed under a heating lamp to activate odorants. Each sample was placed in a 3.3-cm wide tin canister with air holes, then inside a box (25 × 20 × 7.6 cm) that was identical across conditions—but labeled on the bottom—to keep experimenters blind to condition. After each day of experiments, samples were re-frozen, never exceeding 2 hours and 15 minutes at room temperature. Each sample was used for a maximum of 2 days, with no more than two subjects per day.

Testing Room

Trials took place in one of two similar rooms in the Dog Cognition Lab, each with one door and no windows. Room 1 (3.35 × 3.53 m) hosted 31 subject dogs and had a single ceiling vent as the only airflow source. Room 2 (2.77 × 2.81 m) hosted 12 subject dogs and had no identified airflow source. Floors were cleaned with 70% isopropyl alcohol between subjects to minimize the transfer of scent information between the dogs. They had access to water throughout the study and were never separated from their owners. 

Experimental Design

Owners brought their dogs to the Dog Cognition Lab on Barnard College’s campus in New York City with completed consent forms and received study instructions as detailed below. Data collection took place in April 2024, November 2024, and February–March 2025. Owners remained with their dogs throughout the study to promote natural behavior. Sessions were scheduled to avoid overlap between participants. Trials lasted for a maximum of 45 minutes and occurred between 5:30pm and 7:30pm. Dogs were given 5 minutes off-leash to habituate to the room. Owners were unaware of the types of odors and were told they were “natural odors” if asked. Experimenters instructed owners to allow their dogs to behave naturally.

Part 1: Habituation-discrimination paradigm

Each dog participated in three habituation-discrimination trials, where they were tested with one of three odor pairings—stress vs. joy, joy vs. baseline, or baseline vs. stress—assigned in a randomized order. An inter-trial interval of 1 minute separated each trial. Dogs began each trial with a “treat-box” presentation to encourage engagement in the study. The owner and dog started sitting, and an experimenter placed a treat-filled box 2.26m from them, using a standardized script of dog-directed speech and gestures to draw the dog’s attention. The experimenter then rose and moved to the corner, turning her back to the owner and dog. This action and a verbal “okay” signaled to the owner, as pre-instructed, to stand up, step towards the stimulus, and complete a distraction task (i.e., reading a poem). Owner behavior was monitored via video. If the dog consumed or investigated the treat box, odor trials began.

Emotional odor trials followed the habituation-discrimination paradigm and consisted of two phases: a “habituation” phase and a “discrimination” phase. In the habituation phase, the experimenter presented three consecutive, single-box presentations of the same odor (e.g., the joy odor). Following this, in the discrimination phase, the experimenter presented two boxes—both the habituated odor (e.g., the joy odor) and a novel odor associated with a different emotional condition (e.g., the stress odor)—side-by-side. Box side placement was randomized to avoid side bias. Experimenters followed the same script as in treat-box presentations. Emotional odor presentations lasted ≥20 seconds. All trials were videorecorded for later coding of investigation times. 

An “owner-enhanced” protocol was used if dogs failed to approach the odor boxes during the first habituation presentation or during the discrimination phase. In these cases, upon the experimenter presenting the odor(s), turning away, and saying “okay,” the owner was instructed to drop the leash, walk to the experimenter’s previous position, and hold the odor canister above the box(es) at the dog’s eye level. The owners maintained eye contact with the canister for the habituation phase or between the two canisters for the discrimination phase.

Part 2: Testing behavioral reactivity to conspecific emotional odors

Twenty-six dogs continued to Part 2. After a 1-minute break, dogs were off leash and exposed to a single odor assigned randomly: either joy, stress, or control (a cotton pad with no odor added). Nine dogs were exposed to the joy condition, nine dogs to the stress condition, and six dogs to the control condition. The experimenter presented a box with one odor, following the presentation method from Part 1, and then the owner was instructed to walk behind the box, focus their gaze on the canister, and hold up the canister to the dog’s eye-level for ≥20 seconds. The owner and stranger (a second experimenter) then sat in seats on opposite sides of the lab, equidistant from the odor box. Both were blind to the odor condition and occupied with a task (i.e., filling in a survey) to avoid interacting with and influencing the dog. The dog moved freely around the room. Trials ended after 1 minute or 1 minute after the dog first investigated the odor. Two dogs did not approach and were excluded from analyses.  All trials were video recorded for later coding of behavioral frequencies.

Data Analysis

Part 1: Video coding for investigation time (in seconds) was conducted frame-by-frame in BARC v.1.0.0 (https://github.com/emilydringel/BehavioralCodingSoftware). Investigation time was defined as when the dog’s nose was within 10 cm and oriented toward the box. Trials were coded until the odor box was retrieved by the experimenter. The coder was blind to the odor conditions presented. Wilcoxon signed-ranked tests assessed (1) habituation (investigation time of 1st vs 3rd presentations of habituated odor), (2) dishabituation (novel odor vs. 3rd presentation of habituated odor), and (3) discrimination (novel odor vs. 4th presentation of habituated odor). Linear regression analyses tested for side bias during the discrimination phase for each emotional pairing. Data collection time period was included as a covariate in these analyses due to procedural and room differences across months (owner distraction task was implemented in November 2024 and February–March 2025 but not in April 2024, and testing occurred in different rooms between April and November 2024 vs. February–March 2025). 

Part 2: Behavioral frequencies were recorded every 5 seconds (12 points/trial). For stress and joy behaviors, every sample point captured all behaviors exhibited during the 5 seconds leading up to that point. Proximity was scored based on which target (owner, odor box, stranger) the dog was closest to and within one body length of. During the 5-second window before each sample point, each target within this range received a score of 1 while others received 0; multiple targets could be scored per sample. Only one posture (upright, sit, or prone) was coded at the end of each 5-second interval, receiving a score of 1 while other postures received 0. Each behavior's data was averaged across dogs per sample point in each odor condition, and the normality of each behavior’s dataset was assessed using the Shapiro-Wilk test. Differences in behavioral frequencies across conditions were analyzed using ANOVA with Tukey’s pairwise post hoc comparisons for normally distributed data and Kruskal-Wallis test with Dunn’s post-hoc comparisons for not normally distributed data. The effect of the interaction between sex and emotional odor condition on dogs' behavioral responses (stress/joy behaviors and proximity) was analyzed using a generalized linear model (GLM) with a Gaussian distribution. All statistical analyses were performed with RStudio 2023.06.2+561.