To address the uncertainty of the target weight in assessing science and technology awards, the Random Forest Weighting Method was proposed. Reliability analysis was used to calculate the generalization error of expert evaluation data，and according to the minimum error rate, to obtain the evaluation project of each evaluation index weight and to reduce the impact of subjective weighting.Consistently reliable information, inconsistently reliable information and net reliable information were introduced. An improved ELECTRE-Ⅲ method was put forward to solve the problem of expert evaluation data for order variability. A specific number to measure a certain degree of an evaluation project is superior to other evaluation project. Empirical evidence shows that the combination of the Random Forest Weighting Method with the improved ELECTRE-Ⅲ method not only improve the accuracy and reliability of weight estimation , but also solve the problem of inability to set threshold level and rank completely, leading to more scientific, objective and reasonable evaluation results.