This paper presents a new analysis and design method for model reference adaptive control(MRAC) with arbitrary bounded input nonlinearities. The adaptive algorithm ensures that the closed-loop be locally stabilized in the presence of unmodelled dynamic, bounded disturbance and nonlinear input. A modiÞed MRAC scheme, which uses output error to compensate the possible bad performances of the robust adaptive controllers, is proposed to improve the transient and steady state performances, and the ideal asymptotic properties of MRAC is not destroyed with a special feedback compensation.