The fundamental of three major color separation algorithms was analyzed, of which were Neugebauer equations, polynomials, and neural networks. The algorithms were designed with new methods: polynomials were solved by multiple regression analysis; neural networks were modeled by RBF; Neugebauer equation by iteration method. They were programmed by Matlab and experiments were designed to compare the accuracy and stability of the three algorithms. The results showed that polynomial regression algorithm's stability and accuracy were the best and has the smallest average color difference; RBF algorithm was the fastest. Improved method was proposed by analyzing separation results.