Describes 2 experiments with a total of 203 introductory psychology undergraduates. Exp I compared recognition memory for 2 types of homographs. For balanced homographs, there are 2 relatively equiprobable semantic encodings; for polarized homographs, there is 1 dominant encoding. It was predicted, on the basis of encoding variability theory, that recognition would be superior for polarized homographs; however, the opposite outcome was obtained and replicated at 2 retention intervals. In Exp II, successive word association tests were administered at 2 intertest intervals. The semantically scored responses support the assumption of greater variability of semantic encodings for the balanced homographs. A retrieval strategy of semantic recoding is proposed to account for the better recognition of balanced homographs.