In this paper, the multi-model probabilities for skin color detection are proposed to find the regions of skin color in images with human faces. It can deal with skin colors of different races. The captured skin-color information is further applied to determine the health condition of users by comparing with the pre-stored database. The system will be very useful in the integrated surveillance systems, which combined with face detection and face recognition techniques to track any registered user. In the experiments, the testing image database includes 191 image pairs. The results showed that the accuracy of skin color detection is 91.95% and the accuracy of the health condition is 84.8%. Both of them are approaching the status of real applications.