Abstract: This series of studies investigated horses' ability to learn the concept of sameness under several different conditions. Before experimentation began, three horses were shaped to touch individually presented stimuli with their muzzles, and then to make two responses to two matching cards from an array of three. A modified version of the identity matching-to-sample (IMTS) procedure was used to present stimuli in a variety of configural arrangements on a barn wall (Experiment 1 and Experiment 2), and on a flat panel mounted to a barn door (Experiment 3). The task in each experiment was to select the two stimulus cards that were the same (either circles or Xs) and to avoid the nonmatching stimulus card (either a star or a square). In Experiment 1, the mean accuracy rate for selecting the matching alternatives was 74%. The horses' accuracy levels reached a mean level of 83% during Experiment 2, in which they received additional trials and an intermittent secondary reinforcement schedule. In Experiment 3, when the stimuli were moved further apart from each other within arrangements and were presented on a novel background, the mean accuracy rate was 73%. These data demonstrate that horses can learn complex discrimination problems involving the concept of sameness, and that they are able to generalize this learning to a novel stimulus presentation situation. These results also suggest that a relational discrimination test may be useful for assessing horses' learning ability and the level of training appropriate for individual horses.