A new indirect adaptive neural network control scheme with projection algorithm was developed for a class of uncertain nonlinear time-delay systems in this paper. The design was based on the principle of variable structure. Multi-layer Neural Networks (MNNs) were utilized to approximate for unknown plant functions. With the help of a supervisory controller, the resulted closed-loop system was globally stable in the sense that all signals involved were uniformly bounded. Furthermore, the adaptive compensation term of the optimal approximation error was introduced to minimize the effects of modeling error. By theoretical analysis, it is shown that the tracking error converges to zero. Simulation results demonstrate the effectiveness of the approach.