A new input-output approach is presented for optimal regulatory control of systems based on balanced state realization. State-space identification is achieved using input-output data and a balanced realization algorithm. Based on the realized model an optimal controller design is presented to regulate the system outputs at desired levels. The optimal control algorithm is model-predictive in that predictions are made based on measured disturbances. It also compensates for unmeasured disturbances through integral state feedback. This control concept is tested using a continuous stirred-tank chemical reactor model.