In order to meet the requirements of automotive crashworthiness and lightweight, the reliability optimization design of CFRP anti-collision beam was conducted. With the abscissa, thickness and ply angle of the key control points of the CFRP anti-collision beam as the design variables, and the peak impact force as the constraint conditions, an optimization model was built to maximize the specific energy absorption of the anti-collision beam. The Kriging approximation model of the objective function and the constraint function was obtained by fitting the data with the Latin superelevation method. The UGF-direct mapping method was used for the crash reliability optimization design. The results show that when the random variables are not normal and the function is highly nonlinear, the traditional moment method cannot ensure convergence, the Monte Carlo method has the highest accuracy, but the calculation cost is too high. The reliability analysis and optimization using UGF can ensure stable convergence, and the efficiency is also very high while ensure accuracy. At the same time, the UGF method combined with the non-uniform clustering technology further improves the efficiency. Compared with the initial target value, the optimization magnitude obtained by UGF method reaches 21.6%, reaching the expected effect.