Water distribution networks (WDNs) have suffered severe damages under earthquakes. Therefore, it is of great significance to analyze the seismic reliability of WDNs. However, to our best knowledge, few studies have involved the seismic reliability of in-service WDNs. To this end, in this study, a framework for assessing the seismic reliability of in-service WDNs is presented. Firstly, the pipe daily failure probability is predicted accurately via the deep learning algorithm. Secondly, the pipe fragility model based on pipe seismic reliability is improved by coupling with the daily failure probability, which makes the simulation results of the fragility model closer to the actual damage data of pipes under earthquakes. Thirdly, the nodal heads of WDNs are derived through hydraulic analysis of WDNs with leakages. Then, based on Monte-Carlo simulation method, the seismic reliability of WDNs is obtained by counting the number of times the nodal head exceeds the demand head. Finally, the seismic reliability of an actual WDN in China, is assessed in detail as a case study to demonstrate the proposed framework. Results show that nodal seismic reliability can be influenced by service time, pipe roughness, the distance from water plant to the node and the loop configuration.