Abstract: Chiller is a complex system in which the correlation between variables is serious. When a fault occurs, the symptoms and causes of the chiller show diversity, leading to great difficulty in fault diagnosis of the chiller. To reduce the data redundancy and improve the efficiency of the fault diagnosis, a fault diagnosis method for chiller based on independent component analysis (ICA) and least squares support vector machine (LSSVM) was proposed. ICA was used to extract the correlation of variables of the chiller and feature extraction was made which was served as input parameters of LSSVM in order to identify the chiller's fault type. The method was validated by using the laboratory data from the ventilation and air conditioning training platform of a subway station in Beijing and the method was compared with the traditional fault diagnosis method of chiller. Results show that the method is better than the traditional method. It can effectively extract the data from the high-order statistical information and improve the efficiency of fault diagnosis.