Three different theories have been presented to explain how metaphors help in the learning process: concretizing, assimilation, and structurizing. Each of the theories was outlined along with its theoretical and empirical support. These three theories were shown to predict different patterns of inferences and errors in solving problems. In an experiment using metaphors as teaching aids in university lectures in applied statistics, posttests were used to probe the distinct pattern of predictions made by each theory. The results supported structurizing over concretizing in that very general inferences were made by the students taught with metaphors. Structurizing was also favored over assimilation theory because these general inferences were made without the intrusion of technical errors predicted by assimilation theory. Implications of the results for human information processing and education are discussed.