Classical face detection algorithm works on only near frontal faces. Extending it to other poses and in-plane rotated faces require separately trained classifiers which increases both the training and processing time. We solve this instead by developing a reference model that is capable of detecting upright faces in various poses. Then a probabilistic framework is used to estimate occurrence of in-plane rotated faces. Experimental results showed that the proposed approach can achieve face detection accuracy comparable to state-of-the-art approaches but returns more accurate in-plane rotation angle estimation and is much faster. Unlike other approaches, the proposed method is easy to train, requiring only a small number of images and only one manually labeled face image.