THIS STUDY EXAMINES THE DISTRIBUTIONAL VIEW OF key-finding, which holds that listeners identify key by monitoring the distribution of pitch-classes in a piece and comparing this to an ideal distribution for each key. In our experiment, participants judged the key of melodies generated randomly from pitch-class distributions characteristic of tonal music. Slightly more than half of listeners' judgments matched the generating keys, on both the untimed and the timed conditions. While this performance is much better than chance, it also indicates that the distributional view is far from a complete explanation of human key identification. No difference was found between participants with regard to absolute pitch ability, either in the speed or accuracy of their key judgments. Several key-finding models were tested on the melodies to see which yielded the best match to participants' responses.