Storm surge is a serious natural disaster coming from the sea. It is an abnormal sea surface rising caused by strong atmospheric disturbances, such as strong winds and sudden changes in atmospheric pressure. Therefore, accurate prediction of surge deviation is an important task to avoid property losses and to reduce risk caused by typhoon surge. Many conventional numerical methods for typhoon surge prediction have been investigated, but it is still a complex ocean engineering problem. This article applied a neural network combined with harmonic analysis to predict storm surge and surge deviation. The original data from the Suao Harbor station, Taiwan, invaded directly by a typhoon, will be used to verify the present model. Comparisons with two numerical methods, the MIKE 21 model and the U.S. Federal Emergency Management Agency's model, indicate that storm surge and surge deviation can be efficiently predicted using a neural network.