This study investigated the effect of training on perspective-taking with 22 normally developed adults. The perspective-taking task, similar to that used by McHugh, Barnes-Holmes, Barnes-Holmes, Whelan, and Stewart (2007), required the participants read two related statements per trial presented on a computer screen. They pressed one of two keys to indicate if they thought the second statement was true or false. The statements differed along three dimensions, perspective (Self, Other, or Photo), belief (true- or false- belief), and correct response (true or false). Latency to respond, timed from the end of the statement presentation, and accuracy were recorded. A reaction time task, that requiring participants to indicate if a statement ( This is (colour name) ) about a coloured square was true or false, was included to assess the effect of task repetition on response latencies. There were four blocks of reaction time trials alternating with three blocks of perspective-taking trials (Pre-test, Training, and Post-test). During the Training phase there was feedback on the accuracy of each response. Feedback was not given in the Pre-test, Post-test, or reaction time trials. Extended training on the perspective-taking task reduced latencies on this task over and above the decreases seen in the latencies on the reaction time tests, and this reduction generalized to a novel stimulus set. The Self and Other questions resulted in longer latencies than the Photo questions (both before and after the removal of reaction time) as predicted by Relational Frame Theory. The longer latencies were associated with greater relational complexity and partially replicated the results of McHugh et al. (2007a). These results suggest that training with multiple exemplars can be used to decrease response latencies, and so to improving performance, on a perspective-taking task.