Using techniques from evolutionary game theory, we analyze potential games with continuous player sets, a class of games which includes a general model of network congestion as a special case. We concisely characterize both the complete set of Nash equilibria and the set of equilibria which are robust against small disturbances of aggregate behavior. We provide a strong evolutionary justification of why equilibria must arise. We characterize situations in which stable equilibria are socially efficient, and show that in such cases, evolution always increases aggregate efficiency. Applying these results, we construct a parameterized class of congestion tolls under which evolution yields socially optimal play. Finally, we characterize potential games with continuous player sets by establishing that a generalization of these games is precisely the limiting version of finite player potential games (Monderer and Shapley (1996)) which satisfy an anonymity condition.