In this paper, we present a novel method which uses a two-level classifier scheme for eye location. It aims for an efficient eye location process with high environment variance. In this scheme, the image context is organized in the first-level. The second-level classifier performs an actual object detection using two-class discrimination classifier. However, the problem of how to identify the optimal cluster for test images is not yet solved clearly. So we describe a novel method, for first-level getting multiple candidate clusters, and for second-level fusing the outcomes of candidate classifiers which are based on the candidate clusters in first-level. It allows carrying out eye location mission in an optimal way under high environment variance such as illumination intensity, direction, etc. The eye location system achieves the capacity of the high accuracy and change-tolerance by taking advantage of two-level classifier scheme. The experimental results show that the eye location system can achieve superior performance to previously one with the proposed scheme.