Water distribution systems are important lifelines that can be affected by natural disasters, such as earthquakes. It has been reported that it takes one to two weeks or more to restore the water distribution system to a fully functional level after a major earthquake strikes. So, it would be beneficial to society if one could predict the damage∕repair rate experienced by the water distribution system. These predictions can be used by city planners or developers to identify the weak sections in the system and to improve its resilience. In this paper, feed forward back error propagating multilayer neural network is used for the prediction and it is trained using the damage statistics available for different earthquakes in twelve different locations. The training was done using earthquake magnitude, pipe diameter and material type, and peak ground velocity (PGV) as input parameters and the damage∕repair rate per thousand ft as output parameter. The output of the neural network model was compared with two other empirical formulae and the results were found to be promising.