Learning procedures and type of tasks have become useful variables to explain the amount and quality of retention and transfer in problem solving. However, the results are conflicting and often insignificant due to limitations in the design and analysis. The purpose of the present experiment is to examine the effectiveness of several learning methods on transfer and to investigate the possible interaction between learning methods and type of task. A 2 x 4 factorial design involving two different types of tasks (anagrams and Luchins' water-jar problems) and four treatments {both rules and examples given (Condition RE); rules given but without examples (Condition RE); examples only without rules (Condition RE); and neither rules nor examples (Condition RE)} were used. As a response measure we used the percentage of problems solved successfully on posttest and transfer. The results show (1) highly significant and consistent interactions of Tasks x Treatments and Tests x Treatments; (2) powerful influence on both retention (positive efforts) and transfer (negative effects) by the rule factor; (3) overall, better learning resulted from the discovery method. Functional fixedness was hypothesized and confirmed, particularly with anagrams. Several factors responsible for the superiority of discovery over rule learning are presented as well as some implications for developing a more rational technique of instruction