Summary: Model reference adaptive control procedures that do not require explicit parameter identification are considered for large structural systems. Although such applications have been shown to be feasible for multivariable systems, provided there exists a feedback gain matrix which makes the resulting input/output transfer function strictly positive real, it is now shown that this constraint is overly restrictive and that only positive realness is required. Subsequent consideration of a simply supported beam shows that if actuators and sensors are collocated, then the positive realness constraint will be satisfied and the model reference adaptive control will then indeed be suitable for velocity following when only velocity sensors are available and for both position and velocity following when velocity plus scaled position outputs are measured. In both cases, all states, regardless of system dimension, are guaranteed to be stable.