A controlled trial of a probabilistic reasoning curriculum to change test ordering behavior was conducted. Twenty-four medical interns in university, veterans, and community hospitals were randomly assigned to “probabilities” or “placebo” groups. Four months after the curriculum sessions, interns in the probabilities group wrote 16% fewer orders for “little ticket” laboratory tests than placebo group interns (P = 0.032). The probabilities group ordered fewer tests than the placebo group at all hospitals, but differences were greatest in the community hospital. Probabilities interns ordered ten of 97 tests significantly less often in the postcurriculum period than at baseline; multitest panels were particularly affected. Placebo curriculum interns ordered three of 97 tests less often in the postcurriculum period. Differences in test ordering were not explained by differences in case mix. Testing rates correlated slightly with case mix and length of stay at baseline; all three measures correlated strongly in the postcurriculum period. Teaching about probabilistic reasoning improves the efficiency of test ordering.