The operation of a water distribution system is generally assessed using two complementary computer models. First, the pipe velocities are predicted from a hydraulic model with given input parameters and a network graph. The results of this model are then used in a transport reaction solver to determine various indicators for water quality, such as residence times, source tracing, and disinfectant concentrations. Parameter estimation is required to improve the predictability of the two-coupled models because of high parameter uncertainties,. The objective of this presentation is to validate the gradient calibration method previously proposed for the coupled models. Based on field data measurements, such as tank levels, pressures, flow rates and chlorine concentrations, a weighted least-squares problem is defined for an overdetermined system to minimize the residuals between observed and model values. A direct solution based on the Levenberg-Marquardt method is then proposed. Numerical tests were conducted on two real networks and the dual calibration was showed to be effective even with unexpected circumstances. The parameter uncertainty or even the observability, may be strongly linked to meter placement. Selecting meters that are more sensitive may drastically improve the conditioning of the calibration problem.