In order to remarkably improve the spatial resolution of the fused multispectral images and preserve the original multispectral characteristics as much as possible, a selective remote sensing image fusion method is proposed based on the local feature of contourlet coefficients. Firstly, a window neighborhood mobile template is used to calculate the different local features of corresponding contourlet coefficient matrix one by one for the approximate components and the detail components of each direction of each layer resulting from contourlet transform according to the different fusion purposes of low and high frequency parts in the fusion process of multi-spectral and panchromatic images. Then the approximate images and detail images are fused selectively in contourlet coefficients domain by applying different fusion rules based on proper criterion. The resultant image with high resolution and multi-spectral characteristics is obtained by inverse contourlet transform and inverse intensity-hue-saturation (IHS) transform. Landsat TM and SPOT images are used to conduct the fusion experiment and the results show that the proposed algorithm can remarkably enhance the spatial details and well preserve the spectral features of the original images. This algorithm performs better than traditional fusion methods.