Mate-choice imprinting, the determination of mating preferences at an early age based on an individual's observation of adults, plays a role in mate choice in a wide variety of animals. Theoretical work has thus far been focused either on the effects of mate-choice imprinting on the evolution of the male trait used as a mating cue, or on the evolution of imprinting against a nonimprinting background. We ask the question: if multiple types of imprinting are possible in a species, which is likely to evolve? We develop a haploid population genetic model to compare the evolution of three forms of imprinting: paternal, maternal, and oblique (nonparental adult) imprinting. We find that paternal imprinting is the most likely to evolve, whereas maternal and oblique are nearly equivalent. We identify two factors that determine a strategy's success: its "imprinting set," the set of individuals imprinted upon, and phenogenotypic disequilibrium, the association between imprinted preferences and mating cues. We assess the predictive power of these factors, and find that the imprinting set is the primary determinant of a strategy's success. We suggest that the imprinting set concept may be generalized to predict the success of additional imprinting strategies, such as mate-choice copying.