In railway systems, wear-induced wheel flats (out-of-roundness wheel shape) are among the most common local surface defects. They reduce the train ride comfort, increase fatigue risk and raise safety concerns. A good knowledge of wheel flats (presence and position) can help decision makers avoid expensive operating interventions and unnecessary replacement of wheels. Although a significant amount of research has focused on wheel flat detection with the help of various monitoring systems, the quantification of wheel flats without interrupting railway operations is still challenging. Uncertainties are non-negligible in this context. Unfortunately, in most existing wheel-flat-detection methods, uncertainties are rarely taken into account. In this paper, a model-falsification framework is presented to integrate information obtained from measurements with simulation results. Uncertainties associated with both the model and measurements are combined and used to define the criteria to identify flat sizes. The proposed approach has been applied to a field test in Singapore. In the test, rail-pad-force measurement is obtained while a train with a wheel flat is running through the test track. The ranges of values for flat size identified using the proposed methodology include the true observation. It is also shown that the quantification of uncertainties is essential as the predictions that neglect them lead to an underestimation of the flat size.