An original criterion to identify optimum single staee, single phase torque converters geometry is proposed through definition of a number of candidate 'figures of merit', and use of Non-Linear Programming (NLP) techniques to perform their maximization. The mathematical model, inclusive of two-dimensional flow effects through converter bladings, is briefly considered while its implementation into NLP language in terms of an 'objective function' to be maximized subject to a set of constraints is discussed. NLP solutions, based on a FORTRAN IV Code (COOP 77) prepared by the authors, are reported. (A)