The cause of the target fade-out in Neural Network nonuniformity correction scheme for infrared focal plane array(IRFPA)was studied. Based on analyzing the strength and limitation of edge-directed NN scheme(ED-NN-NUC) and nonuniformity correction combining one-point calibration and NN-NUC, a new combinational algorithm for nonuniformity correction was proposed. The new algorithm included three modules which are pre-correction, rough correction and accurate. The comparison experiment with real IRFPA infrared image including dim targets shows that the proposed algorithm can effectively eliminate the targets fade-out and was less time consuming than ED-NN-NUC.