With increasing demands on limited water resources, regulation of larger river systems continues to increase and so too does the need for accurate water accounting and prediction in these systems. River system models are either calibrated manually or automatically on a reach-by-reach basis, i.e. each reach is calibrated as a separate entity with little or no consideration of fluxes at other locations within the river system. While this is a practical approach, simulation errors can propagate downstream to make calibration or prediction difficult at those locations. Likewise parameters may suffer from over-fitting especially where observations are erroneous. We developed and implemented a system calibration strategy in a portion of the Murrumbidgee River, Australia, where parameters for 11 gauges (36 parameters) were calibrated together. Parameter values, model states and model goodness of fit were compared to reach-by-reach calibration. The system calibration produced a better goodness of fit across the whole system relative to reach-by-reach calibration. Additionally, model system states were more realistic than reach-by-reach optimized models. Over-fitting was obvious using the reach-by-reach method for one reach/gauge in particular. This was avoided with system calibration method, with improved goodness of fit at all gauges downstream of the problem gauge. The results here suggest that the system calibration approach provides more hydrologically consistent states, improved overall fit and avoids over-fitting at problem gauges. Copyright © 2015 John Wiley & Sons, Ltd.