This paper develops neural network (NN) models to examine how key design elements of a social website will affect users' feelings or perceptions. An experimental study of 96 university websites is conducted based on a user-centered approach. The study identifies seven website design elements and 33 representative websites as experimental samples for training and testing four NN models. These four NN models are built to formulate the relationship between seven website design elements and three users' feelings of websites. The result of the study shows that the combined NN model has an accuracy rate of 83.93% for predicting the values of three users' feelings of websites. This suggests that the combined NN model is a promising approach for modeling users' specific expectations of websites, thus providing an effective mechanism for facilitating user-centered interface design of social websites.