The impact of inaccurate or unreliable hydrological data on model calibration can be significant. A special case is when calibration data are from different sources with varying accuracy, e.g., missing streamflow measurements that have been supplemented by interpolation or extrapolation. Since the relative accuracy of data is commonly defined subjectively, model calibration with data of varying uncertainty is difficult. In this work, a framework using the method of order of importance is presented to incorporate soft information to describe the relative accuracy of calibration data. A numerical example using synthetic data is presented to demonstrate the validity of the methodology. The applicability of the framework is demonstrated for the Fishtrap Creek Catchment of Washington State, where the short streamflow record with significant gaps was reconstructed using support vector machines (SVM). Incorporation of this educated judgment about the relative accuracy of the calibration data resulted in identification of faulty model calibration as well as errors in the SVM prediction of extreme streamflow events, which would have been undetected otherwise. The methodology was further applied to two management scenarios involving flood frequency analysis and satisfying legally established in-stream flow requirements.