A new robust attitude control design of spacecraft is proposed by combining sliding mode control (SMC) and productive networks (PN). Essentially, the sliding mode control uses discontinuous control action to drive state trajectories toward a specific hyperplane in the state space, and to maintain the state trajectories sliding on the specific hyperplane. This principle provides a guideline to design a robust controller. Productive networks, which are a special type of artificial neural network, are then used to implement reaching and sliding conditions, and tackle the drawbacks of SMC such as chattering and high control gains. Attractive features of the proposed method include a systematic procedure of controller design, a reduction in chattering, robustness against model uncertainties and external disturbances. An inverted pendulum and a spacecraft attitude control problem are given to deomonstrate the effectiveness of the proposed method.