A model reference adaptive-sliding mode control is presented and applied for a variable stiffness actuated (VSA) system. The VSA is a flexible stiffness machine with two coupled actuators in each link, and its safety during the movement can be guaranteed by varying both stiffness and the angular variables. Realisation of precise control of two coupled actuators poses a considerable challenge, however, because of uncertain time-varying parameters and unknown variation bounds. In this paper a neuro-sliding mode approach based on model reference adaptive control (MRAC) is proposed. The proposed MRAC control structure induces the VSA to follow its nominal dynamics with help of sliding mode control efforts. The sliding gain, implemented by a simple neural network (NN), is adaptively updated based on the Lyapunov criterion. A control law and adaptive laws for the sliding mode control as well as the weights in the NN are established so that the closed-loop system is stable in the sense of Lyapunov. The tracking errors of both the angular variables and stiffness are managed to guarantee the system to be asymptotically stable rather than uniformly ultimately bounded. And, the feasibility of the proposed control approach is demonstrated by means of experimental results as well as computer simulations.