La eficiencia de transmisión es una característica clave de la Transmisión Variable Continua Hidromecánica (HMCVT), que está relacionada con el rendimiento de los tractores de servicio pesado. La predicción de la eficiencia de transmisión de HMCVT es beneficiosa para el ajuste en tiempo real de la relación de transmisión durante las operaciones de tractores de servicio pesado, con el fin de obtener un mejor rendimiento. Aiming at the problems of accurate method, low accuracy, and high noise in the prediction of HMCVT transmission efficiency, this paper proposes a method based on Variational Mode Decomposition (VMD), Particle Swarm Optimization (PSO), and Back Propagation (BP) neural networks to improve the quality of transmission efficiency prediction. Firstly, a simple theoretical model was established to obtain the influencing factors of transmission efficiency. Then, based on these factors, the transmission efficiency was tested on the bench under multiple conditions and the influence degree of each factor on transmission efficiency was divided using Partial Least Squares (PLS) method. Finally, the VMD method was used to denoise the test data, and a BP model, which was improved using the PSO method, was established to predict the processed data. The results showed that transmission efficiency of HMCVT is most affected by output speed, followed by power, and least by input speed. The VMD method can accurately extract effective signals and noise signals from the original data, and reconstruct signals, reducing the noise proportion. Using three conditions, the prediction regression accuracy of the PSO-BP model is 7.02%, 7.88%, and 9.26% higher than that of the BP model, respectively. In the three prediction experiments, the maximum differences in the MAE, the MAPE, and the RMSE of the PSO-BP model are 0.002, 0.463%, and 0.004, respectively, which are 0.006, 0.796%, and 0.003 lower than those of the BP model.