We investigated the spontaneous detection of "wrong notes" in a melody that modulated continuously through all 24 major and minor keys. Three variations of the melody were composed, each of which had distributed within it 96 test tones of the same pitch, for example, A 2 . Thus, the test tones would blend into some keys and pop out in others. Participants were not asked to detect or judge specific test tones; rather, they were asked to make a response whenever they heard a note that they thought sounded wrong or out of place. This task enabled us to obtain subjective measures of key membership in a listening situation that approximated a natural musical context. The frequency of observed "wrong-note" responses across keys matched previous tonal hierarchy results obtained using judgments about discrete probes following short contexts. When the test tones were nondiatonic notes in the present context they elicited a response, whereas when the test tones occupied a prominent position in the tonal hierarchy they were not detected. Our findings could also be explained by the relative salience of the test pitch chroma in short-term memory, such that when the test tone belonged to a locally improbable pitch chroma it was more likely to elicit a response. Regardless of whether the local musical context is shaped primarily by "bottom-up" or "topdown" influences, our findings establish a method for estimating the relative salience of individual test events in a continuous melody.