Despite donkeys being involved in various activities with humans, their cognitive and learning abilities are still little known. A deeper understanding of their perceptive, cognitive and learning processes is, thus, necessary to preserve their well-being and establish a good human-animal bond. An operant conditioning task was applied to explore donkeys’ learning abilities. Nine out of 14 adult and non-working donkeys of both sexes fully completed a three-phases training procedure. In the first phase, animals could approach the manipulandum, a specifically designed cabin with a button on the front side. Thereafter, donkeys had to learn how to interact with the button to obtain a feed reward. To evaluate donkeys’ learning capabilities, two linear models were built: i) a fixed-effects model, exploring how the average between-pressures time (BPT) was affected by individual characteristics (i.e. age, sex, and donkeys’ height at the withers) and the training session; and ii) a mixed-effects model to evaluate the difference in the average between-pressures time among consecutive sessions (BPTcs) in the function of the animals’ characteristics, including the sessions-lags as the random effect. In the first model, all the explanatory variables resulted significantly associated with the BPT observed variability. Male donkeys presented a BPT significantly higher, increased by 23.14 s (S.E.=9.71, p = 0.003) compared to females of the same height and age. Age was significant with a positive coefficient (Est.=1.21, S.E.=0.55, p = 0.032). The ‘high’ height class estimate was significant (Est.=13.06, S.E.=6.26, p = 0.032), while no significant effects were identified between the ‘medium’ and ‘short’ and the ‘medium’ and ‘high’ height classes. Lastly, the variable ‘session’ was significant with a negative coefficient (Est.=−10.64, S.E.=1.56, p < 0.0001), indicating an increase in the average speed to perform the desired behaviour for each additional training session. In the second model, the variable ‘sex’ was the only predictor significantly associated with the BPTcs, indicating that the male group progressively improved performance time faster than females (Est.=−8.71, S.E.=4.19. p = 0.045). This pilot study: i) provides insights into donkeys learning abilities by applying an effective methodology for operant conditioning; ii) it highlights how intrinsic animal characteristics might affect asses training performances, although further points need to be explored in future research; iii) it confirms that feed represents an effective positive reinforcement in operant conditioning with donkeys. The development of appropriate handling and training methods, respectful of animals’ subjective experiences and based on positive practices, can improve donkeys’ welfare and their relationship with humans.