Abstarct: In this thesis, neuro-fuzzy-sliding mode controllers are studied for the three-lixnk SCARA robot. The main goals are to design a free model controller and to eliminate the chattering that is produced by the switching term and also to obtain a robust controller against the joint friction and external disturbance. In this method, a neural network used to approximate the equivalent control law and in a same time to evaluate the switching law, a fuzzy system is utilized. The weights of the neural network are updated such that the fuzzy system output (for the multilxayer perceptron network) and the sliding surface (for the radial baxse network) of the neuro-fuzzy-sliding mode controller approaches to zero. The simulation result of this method is compared with conventional sliding mode. Moreover, in order to design a flexible and low cost controller and also to reduce the computational burden, two other controllers are proposed. In the first one, the previous neuro-fuzzy-sliding mode controller is redefined in discrete time domain and is compared with discrete sliding mode. Additionally, the sampling time effects on the closed loop system convergence are discussed. For the second one, to design a faster controller the fuzzy system is neglected and the nonlinear sliding surface is used to compensate fuzzy system specifications. In all of three above cases, presented simulation results demonstrate the performance improvement in each cases. The proposed controllers also are applicable to large class of MIMO systems.