Congestion pricing programs toll only a subset of roads and times, with theoretically ambiguous spillovers. We study New York City’s cordon-based congestion pricing using granular data from Google Maps and a generalized synthetic control design. The toll raised cordon-area speeds by 11%, with little effect on air quality, consumer spending, or foot traffic. Speeds also rose outside the cordon, especially on roads frequently traversed by cordon-bound drivers, making many unpriced trips faster. The magnitudes reflect steep congestion functions, where small volume reductions yield large speed gains. Because unpriced trips vastly outnumber priced ones, their small per-trip gains dominate aggregate driver welfare.