A method for remote sensing image band simulation based on ground-object spectral features study by support vector machine(SVM) was proposed, since the existing simulation methods had difficulties in the spectrum bands extension. By taking spectral features into account as a priori knowledge and according to the study of the complicated and nolinear relationships among ground-object spectra in different bands by SVM, a new band image was simulated on the basis of the existing sensor bands of multi-spectral remote sensing images. Based on TM red band simulation experiments, it is confirmed that the method can precisely simulate the real spectral images and the simulation results are reliable. Furthermore, the method was used to simulating IRS true-color experiments. The results verify that it is a practical method to reconstruct the missing bands of multi-spectral image. The method effectively resolves the band defection problem of multi-spectral images, and it also resolves the spectral bands deficiency of high spatial resolution remote sensing image to some extent. The method puts forward a new idea for establishing links between ground-object spectrum and remote sensing pixel spectrum.