A dual-band infrared image fusion algorithm for human visual systems is put forward. First, the source images are decomposed into the multi-scale and multidirectional subbands by using a nonsubsampled contourlet transform method. Then, the highpass subband is fused by selecting the maximum absolute operator and the lowpass subbands is fused by using a method based on region variance. Finally, the fused image is obtained through image reconfiguration. A fusion experiment is made on the dual-band infrared images collected. The method is compared with the gradient pyramid transform and the wavelet transform in fusion effectiveness. The experimental result shows that this dual-band infrared image fusion algorithm for human visual systems is effective and feasible. It can obtain better fusion effectiveness than the common algorithms.