In order to realize the rapid reliability evaluation of long-life and high reliability products at a normal stress level, the reliability statistical model of Arrhenius—Weibull was established, by using the Weibull distributions products with the consistent failure mechanism and a log-linear acceleration life test model between characteristic life and stress level. The minimum variance of the product median lifetime estimate at normal stress level as optimum object was considered, the number and quantity of test stress level, samples allocation proporation and censoring time at each stress level are determined as design variables, put forward a reliability optimum scheme of time-censoring constant-stress ALT（CSALT） for weibull distributions. The Fisher Information Matrix is obtained for constructing asymptotic confidence intervals for the model parameters. As the example analysis shows that the reliability optimum scheme of CSALT is effective.