The combinational evaluation method which combines TOPSIS with gray correlation degree analysis can improve validity of the evaluation result. In such method, a usual way to assign combinational weight of index is subjective weighting, to assign combinational preference coefficient and identification coefficient is manual weighting. The advantages of combinational evaluation method are not fully mined. Aim at this problem, a method uses projection pursuit algorithm to establish combinational evaluation model is presented. The real coded accelerating genetic algorithm is used to dispose those non-linear problems. The gained index weights are objective. In the same time, a new method of confirms combinational preference coefficients and identification coefficients of gray correlation degree is presented. The example shows the combinational evaluation method based on projection pursuit algorithm is scientific and objective.