In this paper, a mathematical model of irrigation main canals is derived from on-line system identification experiments. This model is obtained experimentally by using a robust system identification procedure. The entire process followed to obtain this model is described in this paper: from the design of the experiments to the model validation. The model takes into account prior physical information. A recursive least-squares parameters estimation algorithm with normalization for robust system identification of main irrigation canals is developed. The validation results show the robustness of the parameters estimation algorithm and justify its use in robust model-based control system design. The results also show accurate water levels prediction in open irrigation main canals with time-varying parameters. The reported results show that robust system identification has an important role to play in management and efficient use of water resources. © 2007 Elsevier Ltd. All rights reserved.