Closed-loop systems have been proofed as adequate for automated drug administration. They provide significant advantages for the patient as well as for the anesthesiologist.In essence three different types of implementing the control unit have been established: The classical linear controller (e. g. PID, adaptive PID), rule-based systems (e. g. fuzzy logic) and model-based controller (e. g. PWPD models). Major drawbacks of these implementations can be found in the limited flexibility or the expendable procedures for determining appropriate rules andlor model parameters.Our novel approach, applying neural networks in a predictive model-based design, is able to overcome the depicted difficulties. The performance has been verified by clinical investigations using the short acting non-depolarizing muscle relaxants mivacurium. The obtained results encourage further research using hypnotic drugs.