How do individuals know whether their preferences are moderate or extreme? Are there differences in the speed at which individuals learn where their preferences lie relative to each other? This paper presents a simple model of learning in which individuals invest effort to learn where their preferences rank in the population through a random matching process. Under certain assumptions on the coarseness of the signals exchanged in this process, I find that individuals whose latent policy positions are more extreme become more certain of where their preferences lie relative to the population faster than others. The implications of these results are discussed with relation to electoral competition, campaign donations, interest group formation, and as a possible microfoundation for preference intensity.