Shape memory alloy (SMA) actuators exhibit hysteresis which is often responsible for position inaccuracy in regulation or tracking tasks and even causes instability in severe cases. In this research, a new approach is proposed to compensate hysteresis in SMA actuators by using a feed forward neural network controller and a sliding-mode based robust controller. The former is used to cancel the hysteresis and the latter compensates uncertainties such as the error in hysteresis cancellation and ensures system's stability. A single wire SMA actuator is used as the controlled object in this research. The experiment shows that the actual displacement of the actuator closely followed that of the desired sinusoid command with a peak-to-peak travel of 8 mm. The maximum error observed was 0.15 mm, which is 2% of the total stroke and the root mean square error was 0.0805 mm. This result demonstrates that the proposed method is very effective.