A novel model reduction methodology is proposed to approximate large-scale nonlinear dynamical systems. The methodology amounts to finding computationally efficient substitute models for the nonlinear subsystems. Model reduction is pursued by viewing the system as a grey-box (or hybrid) model with a mechanistic (white-box) component and an empirical (black-box) component. Before identifying the substitute model, the mechanistic subsystem is reduced by projection using proper orthogonal decomposition. Subsequently, the empirical component is identified by parameter estimation to substitute the nonlinear subsystem. As a consequence, a reduced model with less nonlinear complexity is obtained. An example involving a distributed parameter system is provided.