Low Emission Zones (LEZs) - urban areas that exclude vehicles failing to meet certain environmental standards - have become increasingly prominent. Yet little is known about their optimal design: which vehicles to restrict, on what timetable, and over how large an area. Unlike price-based instruments, LEZs ration access inefficiently: their welfare cost is also borne by inframarginal drivers of restricted vehicles. As a result, welfare is quasi-convex in LEZ size, yielding three bang-bang design principles: (i) the LEZ should be either as large as politically feasible or, if it falls below a minimum scale, not implemented at all; (ii) less polluting vehicles, such as hybrids, should be treated as transition technologies - temporarily exempt, but later subject to the same restrictions as dirtier vehicles, and (iii) these later restrictions, which must be announced in advance, require commitment from the planner. We use data on new-car prices and purchases to assess Madrid's LEZ through the lens of this bang-bang logic. The policy shifted buyers toward cleaner cars, but the surplus lost by inframarginal drivers outweighed the gains from the emissions saved by the switchers.