Experimental studies of memory for tonal sequences have been extended to include well‐known melodies and long‐term memory. W. J. Dowling and D. S. Fujitani [J. Acoust. Soc. Am. 49, 524–531 (1971)] used real melodies in an investigation of interval (magnitude of pitch changes) and contour (direction of pitch changes). The study used the recognition paradigm, one of its advantages being that it enables subjects who have no skill in music production to carry out the tasks. However, it may be that the recognition paradigm is more useful for investigating stimulus variables rather than subject variables. The present study used a sample of professional classical musicians and a selected nonmusicians group. Both groups used an original production method to represent melodies and tonal sequences. The performance of the two groups was compared under conditions of dictation and production, using a contour task and a magnitude‐estimation task. Although the musicians were generally superior, there was little difference between musicians and nonmusicians in terms of perception of melodic contour. However, both groups performed at a much lower level when estimating interval magnitudes, suggesting that the size of pitch intervals is not normally coded in terms of magnitude. This finding is discussed in the light of Dowling and Fujitani’s interpretation; and an alternative ’’matching’’ model is proposed.