We evaluated five competing hypotheses about what predicts romantic interest. Through a half-block quasi-experimental design, a large sample of young adults (i.e., responders; n = 335) viewed videos of opposite-sex persons (i.e., targets) talking about themselves and responders rated the targets' traits and their romantic interest in the target. We tested whether similarity, dissimilarity, or overall trait levels on mate value, physical attractiveness, life history strategy, and the Big-Five personality factors predicted romantic interest at zero acquaintance, and whether sex acted as a moderator. We tested the responders' individual perception of the targets' traits, in addition to the targets' own self-reported trait levels and a consensus rating of the targets made by the responders. We used polynomial regression with response surface analysis within multilevel modeling to test support for each of the hypotheses. Results suggest a large sex difference in trait perception; when women rated men, they agreed in their perception more often than when men rated women. However, as a predictor of romantic interest, there were no sex differences. Only the responders' perception of the targets' physical attractiveness predicted romantic interest; specifically, responders' who rated the targets' physical attractiveness as higher than themselves reported more romantic interest.