In the paper, a cost function of the output error (i.e. the difference between the reference model output and the plant output) and weighting control input is minimised in the model reference adaptive control, without the assumption that the unstable plant zeros must be the zeros of the reference model. There are two reasons for minimising this cost function: first, to avoid the cancellations of stable plant zeros in the poorly damped region by the closed-loop system poles, for the purpose of preventing a large ripple control input [1], and, secondly, to reduce the effect of disturbance by a suitable choice of weighting function to control input [2]. Two simulation examples are presented to show the good behaviour of this algorithm.