A new scheme for image fusion based on expectation maximization is presented in this paper. Image formation model is firstly assumed, and the expectation is employed as objective function. Then the parameters of the model is estimated bymaximizing the objective function. The trae scene can be achieved during the estimation p rocess. The ideal fused image is gained. The experimental results demonstrate by several real image data that the p roposed scheme can achieve better fusion performance than the fusion method based on weighted average and selection on multiscale space.