BP neural network is able to model and predict complex nonlinear relationships by learning and adjusting weight parameters, which shows great potential in performance evaluation. After optimizing the BP neural network model by using the particle swarm algorithm, the article proposes a new model for performance evaluation using the BP neural network as the basic model and conducts an empirical study with data from 10 enterprises. The results of the study show that in the implicit layer node trial-and-error method experiments, the overall error of the model shrinks with the increase in the number of neurons. When the number of neurons is 9, the “V-MSE” is the smallest among all the hidden layers with the value of 0.00051, and the value of R2 is 0.9631, which shows that the model has a good convergence and fitting effect.