The schema theory of learning predicts that varied training in motor learning should give rise to better transfer than specific training. For example, throwing beanbags during practice to targets 5 and 9ft away should better generalize to targets 7 and 11ft away, as compared to only throwing to a target 7ft away. In this study, we tested this prediction in a throwing task, when the pretest, practice, and posttest were all completed within an hour. Participants in the varied group practiced throwing at 5 and 9ft targets, while participants in the specific group practiced throwing at 7ft only. All participants reliably reduced errors from pretest to posttest. The varied group never outperformed the specific group at the 7ft target (the trained target for the specific group). They did not reliably outperform the specific group at 11ft, either. The numerically better performance at 11ft by the varied group was due, as it turned out in a subsequent experiment, to the fact that 11ft was closer to 9ft (one of the two training targets for the varied group) than to 7ft (the training target for the specific group). We conclude that varied training played a very limited role in short-term motor learning.