Shape memory alloy (SMA) is a promising smart metallic material, which has the ability to recover its shape when heated. This characteristic enables SMA to serve as an alternative to replace conventional actuators. However, there also exist hysteresis nonlinearities and parameter uncertainties, which make SMA actuators difficult to model and control. This paper develops a gray-box identification and control approach for SMA actuators. To copy with hysteresis nonlinearities, this plant is described as an Hammerstein model, which is a cascade connection of a nonlinear function followed by a linear sub-system. Then the parameter adaptation is performed based on a robust recursive estimator, and the control law compensates the modeling error through incorporating the unmodeled dynamics estimation. The system stability is ensured and the experimental results verify the effectiveness of the proposed method.