In this paper we study the L1 optimal control problem for linear systems. We will show that by allowing the class of controllers to include nonlinear controllers, we can make the closed-loop L1 norm strictly smaller than we could do using only linear controllers.