Building retrofit is an important strategy to reduce energy consumption and carbon emission in building industry. In order to optimize the design of building envelope retrofit, a multi-objective optimization method combining BP neural network and Monte Carlo-non-dominated ranking genetic algorithm (MC-NSGA Ⅲ) was proposed. The DesignBuilder software was utilized for building performance simulation to obtain sample data. The BP neural network was utilized to establish prediction models between building envelope and building performance. The prediction models were used as the fitness function for each objective. Monte Carlo method was used for uncertainty analysis of crossover and variation probabilities. The MC-NSGA Ⅲ multi-objective optimization model was constructed to obtain the Pareto front. Then ideal point method was utilized to discover the optimal parameters combination. A case study of a school building in China was used to demonstrate the feasibility and effectiveness. The results indicate that the proposed method can find a comprehensive trade-off solution and provide references for building retrofit planning and design.