Water distribution systems of large metropolitan centers contain several hundred nodes and pipes and a number of pumping stations and reservoirs. The size and complexity of these networks require very efficient steady-state solution techniques for implementation on large digital computer systems. A new approach to this problem is proposed based on graph theory, sparsity oriented matrix decomposition techniques, and Newton's method of iterative solution of nonlinear algebraic system of equations. Several large-scale networks were analyzed using a FORTRAN coded program called WATER using minimal computer resources. The results included in the paper clearly demonstrate the computational power of this approach.