A study examined in detail the initial encoding of worked-out examples of mechanics problems by "good" and "poor" students, and their subsequent reliance on examples during problem solving. The subjects, three males and five females, were selected from responses to a university campus advertisement. Six of them were working towards bachelor's degrees with varying majors. Two of the eight students had additional post-graduate training in psychology and none had a college physics course, although seven of the eight had taken physics in high school. Students with a range of grade-point averages were chosen so that learner differences could be examined. Subjects were given instruction in physics and required to demonstrate mastery of basic physical principles. Talk-aloud protocols were employed to examine the way the subjects learned and understood the examples, and then used their understanding to solve problems. The subjects were split into two groups post hoc using a median split on their problem solving success. Results indicated that "good" students tend to study examples and exercises by explaining and providing justifications for each action and relate their explanations to the principles and concepts in the text. "Poor" students do not often explain the examples or exercises to themselves, and when they do, their explanations do not seem to connect their understanding with the principles and concepts in the text. The results provide, at a gross level, empirical evidence to support existing artificial intelligence models of explanation-based generalizations. (Seven figures and seven tables of data are included and 27 references are appended.) (RS)