In 2 experiments and a rating study we examined asymmetric transfer of repetition benefit between two easily readable surface forms. Typed second occurrences showed benefit that was similar for words and pseudowords and did not depend on the surface form of the first occurrence. Handwritten second occurrences showed benefit only for words, and benefit was larger when first occurrences were handwritten than when typed. This pattern, which was unrelated to explicit memory, characterized both naming and lexical decision, and benefit transferred between tasks. These results tax current episodic accounts of repetition benefit that are based on retrieving perceptual records or conceptual interpretations, and they tax strongly abstractionist accounts that are based on extremely general visual analysis. A weakly abstractionist account is described, which is based on ease of retrieving category prototypes by instances of varying typicality.