An analysis method about the correlation of loads to wide-area dynamic characters of power system is presented. The generator relative angle is regarded as wide-area dynamic character of a power system. The characteristic vector comprising the operation conditions of the power system and the correlation indices of generator angles is constructed firstly. The load time-variation character is considered in the characteristic vector. The dissimilarity of characteristic vectors is set as the criterion that characteristic vectors are classified by the Chameleon clustering algorithm. Through frequent pattern-based mining of classification results， the support and the confidence of the corresponding association rules can be obtained and the correlations of loads to generator relative angles are determined finally. The study results of a 10-machine 39-bus system indicate that the analysis results are not influenced obviously by the load time-variation characteristics. The analysis method is effective and robust， which will be important to load modeling based on wide-area measurement in the future.