Optimal design of water distribution networks, which are governed by a series of linear and nonlinear equations, has been extensively studied in the past decades. Due to their NP-hardness, methods to solve the optimization problem have changed from traditional mathematical programming to modern intelligent optimization techniques. In this study, with respect to the model formulation, we have demonstrated that the network system can be reduced to the dimensionality of the number of closed simple loops or required independent paths, and the reduced nonlinear system can be solved efficiently by the Newton-Raphson method. Regarding the optimization technique, a discrete state transition algorithm (STA) is introduced to solve several cases of water distribution networks. In discrete STA, there exist four basic intelligent operators, namely, swap, shift, symmetry and substitute as well as the "risk and restore in probability" strategy. Firstly, we focus on a parametric study of the restore probability $p_1$ and risk probability $p_2$. To effectively deal with the head pressure constraints, we then investigate the effect of penalty coefficient and search enforcement on the performance of the algorithm. Based on the experience gained from the training of the Two-Loop network problem, the discrete STA has successfully achieved the best known solutions for the Hanoi and New York problems. A detailed comparison of our results with those gained by other algorithms is also presented.