One of the main problems when developing an eye detection and tracking system is to build a robust eye classifier that can detect the true eye patterns in complex scenes. This classi fication task is very challenging as the eye can appear in different locations with varying orientations and scales. Furthermore, the eye pattern varies intrinsically between ethnic groups, and with age and gender of a person. To cope better with these variations, we propose to use a bio-inspired convolutional neural network, based on the mechanism of shunting inhibition, for the detection of eye patterns in unconstrained environments. A learning algorithm is developed for the proposed neural network. Experimental results show that such network has the builtin invariant knowledge and the discriminatory power to classify input regions into eye and non-eye patterns. A classification rate of 99% is achieved by a three layer network with input size of 32 x 32 pixels.