Abstract: In order to consider the uncertainty of seismic damages and post-earthquake restoration of water distribution pipelines, the Monte Carlo simulation (MCS) is generally used for seismic performance and resilience evaluation of water distribution network (WDN). Due to the large number of damage samples generated by MCS and the long time taken to perform hydraulic simulation of the WDN in each damage sample, the MCS approach usually accomplished with large computational burden. This study utilized the LightGBM algorithm to establish a surrogate model to predict the seismic performance of the WDN in each damage sample, instead of the hydraulic model-based analysis, aiming to improve the computational efficiency of seismic performance evaluation of WDN based on MCS. The inputs of the surrogate model are the hydraulic importance index and the earthquake-induced leakage area of the WDN pipelines, and the outputs of the surrogate model includes the water service satisfaction ratio of user nodes and the whole WDN. The effectiveness of the surrogate model based on LightGBM was verified through a benchmark WDN case, and the relative error of the seismic performance of the WDN evaluated by the surrogate model ranges from 1.09% to 1.88%, whereas the computational time was reduced by 73.1% to 92.3% compared with the traditional hydraulic model-based analysis. For the performance evaluation during the post-earthquake recovery process and resilience evaluation of the WDN, the resilience indexes evaluated by the surrogate model have a relative error of less than 1% compared with the results of the hydraulic analysis model. Therefore, the LightGBM-based surrogate model for the WDN performance evaluation provides an efficient tool for seismic performance evaluation, recovery simulation and resilience evaluation of the WDN.