Automatic detection of human eyes in images plays an important role in face recognition and human computer interaction. In this paper, we present a reliable and fast eye detection method. First, it detects a face region from a gray image using an Ada-boosting algorithm. Second, it estimates the head tilt angle and the vertical location of two eyes using intensity valley and edge patterns. Third, it detects the candidate regions for irises with a simple effective filter responding to the generic eye pattern. Finally it accurately localizes the eye using a fast deformable template matching algorithm.