Longitudinal data from The Rand Corporation's Health Insurance Experiment were used to test the hypothesis that provider continuity can be modeled as one behavioral consequence of patient satisfaction. Bivariate and multivariate analyses (controlling for sociodemographic characteristics, prior use of services, health status, and health insurance plan) supported our hypotheses. A multivariate linear probability function indicated that a 1-point decrease on a general satisfaction scale was associated with a 3.4 percentage-point increase in the probability of provider change. The relationship between satisfaction scores and continuity during the following year appears to be roughly linear; we observed no "threshold" satisfaction level at which the probability of provider change increased markedly. We discuss needed improvements in the measurement of provider continuity and the need for further study of other behavioral consequences of patient satisfaction.