In this paper, the approach of generalized predictive control (GPC) is applied to a multivariable model of the Dinorwig pumped-storage hydroelectric power station. The response of the system with constrained predictive control is compared with the classic PI controller as currently implemented. The results show that GPC offers significantly better performance across the plant's operating range. It is shown that fixed-parameter, constrained GPC produces a faster primary response when the station is operating with a single unit while preserving stability as the operating conditions change when multiple units are online.