Musically trained and untrained children between 4 and 6 years of age were tested for their discrimination of transformations of an unfamiliar six-tone melody. Transformations involved a change in one, two, or all three of the following features: melodic contour, musical intervals, or individual frequencies. In order to manipulate task difficulty, melodies were played at different presentation rates. Results revealed that discrimination performance varied as a function of (1) musical training, (2) what features of a melody were changed, (3) number of features of a melody that changed, and (4) rate of presentation. Musically trained children performed better than children without training, they showed enhanced sensitivity to the more specific melodic features (i. e., individual frequencies), they were better at detecting transformations involving changes in a fewer number of features, and their performance was unaffected by rate of presentation. In contrast, children without training attended primarily to more general pattern features (i. e., melodic contour) and needed a change involving a larger number of musical features for reliable discrimination of a transformation. In general, as the rate of presentation of the melody increased a decrement in performance occurred for untrained listeners, but the magnitude of these effects varied with transformation type.