The implication-realization model's description of tone-to-tone expectancies for continuations of melodies was examined. The model's predictions for expectancies are described with a small number of principles specified precisely in terms of interval size and direction of pitch. These principles were quantified and used to predict the data from three experiments in which listeners were required to judge how well individual test tones continued melodic fragments. The model successfully predicted listeners' judgments across different musical styles (British and Chinese folk songs and Webern Lieder), regardless of the extent of listeners' musical training (Experiments 1 and 2) or whether they were born and raised in China or the U.S.A. (Experiment 3). For each experiment, however, the collinearity of the model's predictors indicated that a simplified version of the model might predict the data equally well. Indeed, a revised and simplified model did not result in a loss of predictive power for any of the three experiments. Convergent evidence was provided in a reanalysis of data reported by Carlsen (1981) and Unyk and Carlsen (1987), whose listeners were required to sing continuations to two-tone stimuli. Thus, these findings indicate that the implication-realization model is over-specified. The consistency that was found across experimental tasks, musical styles, and listeners raises the possibility, however, that the revised version of the model may withstand the original model's claims of universality.