An extended method for the design of discrete-time multivariable adaptive control is presented. Unlike most adaptive control algorithms, future control inputs are, in this method, predetermined by using the long-term predictor. With this concept, applications of discrete-time multivariable adaptive controls, which have been applied to only a limited class of practical problems owing to the unboundedness of the control inputs, can be extended to some types of non-minimum phase process. In addition, it is shown that a more generalized control objective, such as minimizing a sum of quadratic errors plus a penalty on control action, can be easily adopted in the design of adaptive controllers by using the concept of a long-term predictor.