To get the global optimal point, propose an optimize BP neural network by an improved particle swarm optimization (PSO) to implement nuclide identification. It changes inertia weight and learning factor dynamically with self-adaption to optimize the weight value and threshold value of BP neural network. It gets the global optimal value of the particle swarm by training BP neural network to identify models. Finally, it implements nuclide identification by using the optimal weight and threshold value. The experiment shows our proposed method can not only converge to the optimal value faster but also do a good balance between local search and global search. Therefore, it significantly improves the convergence speed and the accuracy of nuclide identification.