In model reference adaptive control systems, because the plant is assumed to be unknown, it is a difficult task to choose a suitable reference model such that the perfect model matching condition (i.e., the plant zeros in |z|≥1 must be the zeros of the reference model) is assured in every adaptive step. In the present paper, an adaptive controller, achieved by minimizing theH2‐norm of the error transfer function between the closed‐loop system and the reference model using the technique of calculus of variations, is developed. It guarantees the stability of a closed‐loop system regardless of whether or not the controlled plant is unstable or nonminimum phase.