energy saving potential of latent heat storage unit was studied(LHSU) by using neural network. Experiments were conducted in an enthalpy difference laboratory to simulate the performance of LHSU. Different neural network structures were built to predict the influence of nonlinear factors including air flow rate, air temperature. Coefficient of performance (COP), energy storage and discharge rates on the unit’s operation performance. Results showed that the predictions agreed well with the experimental data with correlation coefficients in the range of 0.99~1.00, mean relative errors below 2% and very low root mean square errors.