Image fusion based on the linear relationship between panchromatic and multispectral data is feasible and widely used for the fusion of remotely sensed images. The GS method is analyzed by using a GCOS model. It is found that the GS method can estimate the linear relationship through least squared regression. Since the least absolute deviation is more robust than the least square, it is expected that the fusion method based on the least absolute deviation is better than the least square in performance. A novel image fusion method based on the least absolute deviation is proposed and is tested for the IKONOS images. The experimental result shows that this novel method is an effective modification for the common GS method.