Due to computational burden and dynamic uncertainty, the classical model-based control approaches are hard to be implemented in the multivariable robotic systems. In this paper, a model-free fuzzy sliding mode control based on neural network is proposed. In classical sliding mode controllers, system dynamics and system parameters are required to compute the equivalent control. In Radial Basis Function Neural Network (RBFNN) based fuzzy sliding mode control, a RBFNN is developed to mimic the equivalent control law in the Sliding Mode Control (SMC). The weights of the RBFNN are changed for the system state to hit the sliding surface and slide along it with an adaptive algorithm. The initial weights of the RBFNN are set to zero and then tuned online, no supervised learning procedures are needed. In the proposed method, by introducing the fuzzy concept to the sliding mode and fuzzifying the sliding surface, the chattering can be alleviated. The proposed method is implemented on industrial robot (Manutec-r15) and compared with a PID controller. Experimental studies carried out have shown that this approach is a good candidate for trajectory tracking applications of industrial robot.