In this paper, we consider the class of adaptive filters with error nonlinearities. In particular, we derive an expression for the optimum nonlinearity that minimizes the steady-state error and attains the limit mandated by the Cramer-Rae bound of the underlying estimation process.