Little work has studied achievement goals in social interaction situations. The present experiment aimed at contributing to this matter by showing the potential of social interaction (in particular disagreement) to moderate the effects of achievement goals on learning. Participants were led to think they interacted with a partner, sharing opinions about a text that they were studying. Mastery and performance goals were manipulated. During the “interaction,” they received either disagreement or agreement from this bogus partner. Results showed that a condition in which mastery goals were induced led to better learning than a performance goal condition only when the partner disagreed. No differences between goal conditions were observed when the partner agreed. Implications for achievement goal research are discussed.