Asset pricing research provides evidence on the cross-section of asset prices, while it is relatively silent on the potential consequences of the pricing models' postulated differences in expected returns. At the same time, return differences may also result from investor behavior that is characterized by focus of attention on asset subclasses, sectors or even single issues, a phenomenon which we denote as "investor crowding". Realized return differences affect weights and may induce crowding in asset portfolios. We argue that crowding can be costly to investors as it affects the risk-return efficiency of market proxies, potentially causing a negative externality for passive investors. The absence of crowding in a long-run equilibrium can be characterized by a weights martingale condition, which imposes restrictions on risk premia dynamics. As such, static cross-sectional premia are difficult to reconcile with long-run absence of crowding. Suggesting an entropy-based measure of investor crowding, we study observed crowding behavior for the U.S. equity market. Our empirical results reveal that episodes of intense investor crowding are recurrent and that the relation with uncertainty is state-dependent. Uncertainty increases moderate crowding levels, i.e. investors tend to crowd when they are fearful. However, once crowding levels become elevated, uncertainty induces rebalancing away from dominant positions.