In this paper, a methodology is suggested to evaluate the goodness-of-fit between observed and numerically predicted variables in curved open channel flows. The mathematical model uses the continuity and the Reynolds averaged Navier–Stokes (RANS) equations with the RNG κ-ɛ turbulence model, numerically solved with the commercial software Ansys Inc. The concept of design of experiments (DOE) together with some criteria of goodness-of-fit was used in a new coherent suggested methodology for the calibration and validation of the numerical model. The goodness-of-fit criterion was applied to secondary flow variables like helicity, intensity of the secondary current and the vortex index instead of the primary variables like the velocity field and the free surface. This methodology guarantees that the numerical model reproduces values of the response variables very near to the observed values and it is valid for the mesh independence analysis and the setup of other numerical parameters, including those related to the boundary conditions.