Problem solving by reference to rules or previous episodes: the effects of organized training, analogical models, and subsequent complexity of experience.

Subjects learned a microcomputer drawing package under different conditions of training organization and practice complexity. Training instructions were presented in either a random or an organized order, and with or without an analogical model of the software package. Practice trial varied in visual and logical complexity. Performance on paper-and-pencil and problem-solving tests was better following the model than following the no-model condition when practice trials were logically complex; the reverse was true when they were logically simple. Performance on the test of problem solving was also better following organized training than following randomly ordered training when practice trials were visually complex; the reverse was true following visually simple practice. We propose that the subjects performed the tasks by engaging in either episode-based or rule-based processing, and that performance was optimized when the processing used at encoding and retrieval was the same. The acquisition of skill in solving real problems is explained as procedural compilation.