Under some conditions, substituting worked examples for problems or exercises enhances learning and subsequent problem solving. Under other conditions, worked examples are no more effective, and possibly less effective, than solving problems. Using cognitive load theory, we hypothesize that the critical factors for enhanced learning are whether the worked examples can direct attention appropriately and reduce cognitive load. It is suggested that worked examples requiring students to mentally integrate multiple sources of information are not effective because they fail with respect to both of these factors. The results of five experiments using geometric optics and kinematics under classroom conditions provided evidence for these hypotheses. Worked examples, formatted to reduce the need for students to mentally integrate multiple sources of information, resulted in test performance superior to either conventional problems or to worked examples requiring students to split their attention between, for example, text and equations or text and diagrams. We conclude that because traditional worked example formats frequently are random with respect to cognitive factors, they may be ineffective in some areas and require restructuring. © 1990, Taylor & Francis Group, LLC. All rights reserved.