This paper focuses on the effects of light condition, glass-wearing on driver's eye, and proposes a way of human eye detection by Hough transform and neural network classifier. Firstly, two eye candidate regions were selected based on the geometry and symmetry of the iris. Then, coarse human eye positioning is conducted by edge detection and MAE operator. At last, the B-P neural network was utilized to pinpoint the human eye. For six groups of video images captured in three different situations, that is, different light, different backgrounds and different skin color, three groups were performed simulation experiment using matlab7.0. Results show that the algorithm for complex situations human eye detection has a strong robustness, improve the accuracy of eye detection greatly.