To reproduce multi-spectral images on a printer, a new spectral-based color separation method using the nonlinear optimization technology was presented, which could improve spectral and the colorimetric precision simultaneously. Based on the color principles of printer systems and the Neugebauer model, the effect of spectral and colorimetric characteristics of images on printing output was analyzed. This led to the conclusion that spectral and colorimetric errors were both important for printed images. For this reason, the spectral error and colorimetric error objective functions were designed firstly, one presenting the colorimetric precision of the printed images and another for spectral precision. Furthermore, the effective area coverage of the printer primary inks was limited according to the actual printing process. Finally the nonlinear optimization method was used, by which the spectral reflectance of the multi-spectral images could be transformed to the effective area coverage of the inks and the color separation was realized. Experiments show that the proposed method improves the spectral and colorimetric precision simultaneously. The output spectral precision is about 3 to 10 times higher than that of the regressive method only based on the spectral error.