We build a model of mutual fund competition in which a fraction of investors ("unsophisticated")exhibit a preference for familiarity. Funds differ both in their quality and their visibility: Whileunsophisticated investors have varying degrees of familiarity with respect to more visible funds,they avoid low-visibility funds altogether. In equilibrium, bad low-visibility funds are drivenout of the market of sophisticated investors by good low-visibility funds. High-visibility fundsdo not engage in competition for sophisticated investors either, and choose instead, to cater tounsophisticated investors. If familiarity bias is high enough, bad funds survive competition fromhigher quality funds despite offering lower after-fee performance. Our model can thus shed lighton the persistence of underperforming funds. But it also delivers a completely new prediction:Persistent differences in performance should be observed among more visible funds but not inthe more competitive low-visibility segment of the market. Using data on US domestic equityfunds, we nd strong evidence supporting this prediction. While performance differences surviveat least one year for the whole sample, they vanish within the year for low-visibility funds. Theseresults are not explained by differences in persistence due to fund size or investment category.The evidence also suggests that differences in persistence are not the consequence of other formsof segmentation on the basis of investor type (retail or institutional) or the distribution channel.