Bottom-up principles of melodic implication (Narmour, 1990) were evaluated in a melody-completion task. One hundred subjects (50 low training; 50 high training in music) were presented each of eight melodic intervals. For each interval, the subjects were asked to compose a short melody on a piano keyboard, treating the interval provided as the first two notes of the melody. For each melody the first response - the note immediately following the initial interval - was analyzed. Multinomial log linear analyses were conducted to assess the extent to which responses could be predicted by Narmour's (1990, 1992) bottom-up principles. Support was found for all of Narmour's principles, and two additional predictors based on implied tonal structure. Responses of low- and high-training groups were similar.