Predicting the trajectory of a spinning ping pong ball can improve the effectiveness of a ping pong robot in daily training. In this study, the Vicon system was used to capture three-dimensional coordinates of the spinning ping pong ball during flight. Then, a long short-term memory (LSTM) neural network algorithm was improved by combining an adap- tive particle swarm optimization (APSO) algorithm and the attention mechanism, and the APSO-LSTM-attention method was obtained for predicting the trajectory of the spinning ping pong ball. It was found through experiments that the APSO-LSTM-attention method had average displacement errors of 6.01mm, 11.26mm, and 8.97mm in the X, Y and Z axes, respectively, and the final point displacement errors were 15.64mm, 17.93mm, and 11.26mm, respectively, indicating that the method outperformed methods such as recurrent neural networks. The time required to predict the complete trajectory by the APSO-LSTM- -attention method was also short, only 0.0186 s. The results demonstrate reliability of the proposed method in predicting the trajectory of the spinning ping pong ball and its potential application in practical scenarios.