AbstractImproving the applicability of electrical machines depends on knowing their performance ondifferent operation conditions. In this paper a technique based on B-spline neural networks forobtaining a high performance of direct current shunt motors is proposed. This algorithm setsthe control signal on line without the need to know a system model and, therefore, theirperformance is not dependent on the equilibrium point of design and prior knowledge of theparameters. Motor operation is subjected to highly demanding conditions for variant speedreference, also takes advantage of the feature of including a load torque from zero to full withminimal impact on the rotor speed. Time domain simulations and laboratory measurements ina test direct current shunt motor demonstrate the applicability of the proposal.