The popular Generalised Predictive Control (GPC) algorithm is used to derive a discrete-time nonlinear predictive controller for muscle relaxant anaesthesia, in a Wiener structure form, and which exhibits strong nonlinearities as well as varying dead-time and dynamics due to patient-to-patient variability. The proposed controller minimises the GPC quadratic performance index taking into account the nonlinear output predictions obtained using linear predictions of the linear part inferred using the inverse of the nonlinear function. The minimisation problem is solved using numerical search methods such as Genetic Algorithms. Moreover, variations in the model dynamics are estimated using a Recursive Least-Squares estimation algorithm (RLS). The proposed scheme is shown to be superior to the linear GPC algorithm even in the case of a significant mismatch between the assumed model nonlinearity and that in the actual process.