Studies motivated by cognitive load theory experiments (e.g., Ward & Sweller, 1990) suggest that problem solving can interfere with learning, and show that alternative exercises such as studying worked examples often lead to better performance. However, problem solving at has often been found to have benefits for both retention and transfer (e.g., Jacoby, 1978) and some studies (e.g., Charney & Reder, 1986) show better learning from problem solving than from worked examples. Many factors may affect the outcome of such studies. Often the methods that produce difficulties during training seem to lead to the best long term performance (Schmidt & Bjork, 1992) Some evidence suggests that the least capable learners benefit most from relatively active, unstructured training methods (e.g., Hermann, 1969; Whitener, 1989) but others hold the opposite opinion (e.g., Snow, 1989). People often fail to make use of examples unless stimulated, for example, by provision of multiple, varied examples (e.g., Gick & Holyoak, 1983). An experiment comparing three approaches to instruction was conducted in an effort to clarify the circumstances favoring problem solving or study of examples in training. Subjects either first attempted to solve a problem then studied a related example, studied an example then attempted to solve a related problem, or studied a pair of related examples. Problem solving first resulted in slower and less accurate performance during training, as did studying less similar pairs of exercises. Subjects who studied worked examples without problems performed poorly on an immediate test, and the least capable among them had particular difficulty with the most complex tasks. On a delayed test, among subjects who studied the less similar pairs of exercises, those trained with pairs of worked examples performed more poorly on difficult problem components than did problem-first subjects. On the delayed test, among trainees of average or lower intelligence, those trained with problems first were quickest. Consistent with the ideas of Schmidt and Bjork (1992), conditions that produced difficulties during training produced the best ultimate performance. Contrary to Snow's (1989) conclusion, this seemed to be particularly important for the least capable trainees.