This study proposes an identification method ofmulti-source dynamic loads based on independent component analysis, in orderto detect the wave pattern of the load subjected to a structural systemwhose configurational information is totally unknown. The method is based onthe principle that the response of structure is a convolution of load andcorresponding structure impulse response; also, it assumes that individualload sources are statistically independent to each other. When compared withthe existing load identification methods, the features of the methoddeveloped in this study embody three merits: (1) The structuralconfiguration, such as distribution of structural mass, structural stiffnessand damping ratio, could be unknown, and the independence among the actualloads is taken as the goal of optimization; (2) The independence ofindividual identified loads is measured using their mutual information, andthe correlation at different orders is eliminated by the use of gradientdegressive algorithm, so that the independence among detected loads isguaranteed; (3) Loads could be identified from the standpoint of wavepatterns. Numerical simulation indicates that the algorithm is robust onmeasuring point, noise, structure and input load. By normalization, thevalue of the correlation coefficient between identified and actual loads isapproximately 1. Therefore, this method successfully integrates the existingknowledge on load into the load detection algorithm, and could make aconfident estimation of the actual value, which contributes in the practicalsignificance of the developed method.