Past problem–solving research has provided a basis for helping students structure their knowledge and apply appropriate problem–solving strategies to solve problems for which their knowledge (or mental models) of scientific phenomena is adequate (model–using problem solving). This research examines how problem solving in the domain of Mendelian genetics proceeds in situations where solvers' mental models are insufficient to solve problems at hand (model–revising problem solving). Such situations require solvers to use existing models to recognize anomalous data and to revise those models to accommodate the data. The study was conducted in the context of 9–week high school genetics course and addressed: the heuristics charactenstic of successful model–revising problem solving: the nature of the model revisions, made by students as well as the nature of model development across problem types; and the basis upon which solvers decide that a revised model is sufficient (that t has both predictive and explanatory power). © 1995 John Wiley & Sons, Inc.