This paper describes a method for multi-axle moving train loads identification based on simulated annealing genetic algorithm by minimizing the errors between the measured displacements and the reconstructed displacements from the identified moving loads. Experimental studies were carried out to investigate the effect of the proposed method on moving loads identification. A simply supported steel beam model and a model train with three carriages were constructed in laboratory. A series of comparative researches for moving loads identification have been conducted. Effects of moving speed and measurement station numbers on the accuracy of the proposed method are investigated. The results show that the proposed method is accurate and feasible for multi-axle moving train loads identification.