An infrared multi-target tracking algorithm is proposed. The infrared target characteristic value after fusion is regarded as the eigenvector,then combining with the Kalman filter,predicted location of the target in the next moment is given,and search area that takes the predicted location as the center is obtained. All feature matching values of targets are calculated in the searching area. The closest matching values are the target. The simulation results show that the presented algorithm runs quickly and has good instantaneity,and it can eliminate the false targets and can track the multi-target stably.