This paper describes a new face detection algorithm that retains its performance regardless of variations in illumination. It is a hybrid model that performs skin segmentation through a fast but coarse procedure and an adaptive thresholding process. Logical pixel combination of the results is performed to give a single skin-segmented image, which is then tested for the presence of a face through template matching. The model performed well for face detection especially for detecting dark faces under dark illumination. 83.8% and 76.4% detection rates were achieved when the algorithm was tested with images taken under good and poor illumination respectively.