Road traffic leads to an externality: drivers do not account for the time cost they impose on others. In this paper, I study the potential gains from optimal congestion pricing. I develop an urban general equilibrium model which features residential and workplace location, travel mode, and route choices with congestion. The attractiveness of workplaces and residences is also determined endogenously. I provide conditions for the uniqueness of both the competitive equilibrium and the first best planner’s problem and characterize the tax instruments needed to decentralize it. I show how the model can be solved with arbitrary taxes, including congestion toll zones of the kind often implemented in practice. I estimate the model's parameters in an application to New York City. I find that the first best tax policy would realize gains of $0.77 per person per day or a total of $21.7 million per week. Over a third of the gains from optimal congestion pricing at the individual link level can be achieved by a congestion zone that covers only lower Manhattan. I decompose these gains along different margins of adjustment, finding that mode choice is a key driver of the results with driver route choice and general equilibrium location choices also playing a non-negligible role.