The emergence of urban big data is transforming the existing research paradigms in urban studies. New theories and analytical methods are required to meet the methodological challenges. This paper empirically compares a data-driven approach and an urban-system-model approach through a case study of modelling the commuting patterns in Beijing. For the data-driven approach, the novel location-based-services (LBS) data are explored to identify the employment-residence location of the service users. For the modelling approach, a spatial equilibrium model is calibrated for base year 2010 and is used to simulate the commuting patterns for Beijing 2015 based on exogenous development projections. The results of the two approaches are then compared against the benchmark statistics for Beijing 2015. The comparison shows that the LBS data perform better in detecting residence locations than employment locations. The model prediction fits better with the benchmark, while the errors of the LBS data tend to vary significantly across space. For amplifying the LBS sample data to represent the full population, uniform scale factor thus should be avoided. In addition, the ineffectiveness of representing short-distance commuting for the LBS data is revealed by the comparison with the model predicted flows. In light of the strength and weakness of the respective approach, the prospect of a collaborative use of big data and urban system models is explored in the conclusion.