Skin pixel detection technique is a basic and important part of image recognition system related to human being, such as adult image recognition, face recognition, etc. In order to improve the precision of skin pixel detection technique, a skin pixel detection algorithm that combines fuzzy theory with Forward Propagation(FP) Neural Network is proposed. The algorithm composes a feature vector based on extracted color features of the pixel through fuzzy theory and intuitionistic fuzzy theory. The feature vector includes membership and hesitancy degree of the pixel to common skin pixels. Roughness is a supplementary feature in the feature vector in order to completely express the feature of skin pixel. Then the algorithm is used to train a FP neural network and the feature vectors are classified into skin pixels and non-skin pixels. Experiment shows that the algorithm could improve the precision of skin detection and be effectively used in recognition systems related to human being.