Students of reasoning have long tried to understand how people revise systems of beliefs. This paper maintains that people often change their beliefs in ways driven by considerations of explanatory coherence. In this report, a computational model is described of how experimental subjects revise their naive beliefs about physical motion. First, instances in which subjects changed their beliefs while learning elementary physics are presented. Each of these cases involved an individual's attempt to explain a surprising observation. Next it is shown how their belief revisions can be modeled using ECHO, a connectionist computer program that uses constraint-satisfaction techniques to implement a theory of explanatory coherence. The resulting simulations even captured temporal characteristics of the observed changes in beliefs. Finally, the model's representational sensitivity and procedural robustness are discussed. The paper concludes by showing how ECHO can be used to generate empirical predictions about subjects' current beliefs. (ML)