Thermal energy demand represents over 30% of overall energy consumption and thus is a vital target for the decarbonisation of our energy supply. This fact is the reason why dynamic thermal modelling of buildings, components and infrastructure has been gaining increasing importance in the context of the transformation of existing energy systems into smart energy systems. For the first time, this study presents a comparison of two major bottom-up approaches to modelling urban neighbourhoods: a physical modelling approach and a data-driven approach. The physical method is represented by a GIS-based automated modelling approach with detailed dynamic building simulation in IDA ICE. In the data-driven approach, an energy signature is applied, which uses a non-linear data-driven method. These two methods were validated on the basis of a multi-family house, an office building and a residential area with 34 buildings by means of detailed multi-zone building simulation and measurement data. The simulation results show that both applied approaches are applicable in these cases within good agreement compared to the measurement data (physical/data-driven: multi-family house (RN_RMSE(%) = 14.7/9.3; R2 = 0.68/0.87), office building (RN_RMSE(%) = 7.6/5.4; R2 = 0.92/0.96) and residential area (RN_RMSE(%) = 8.2/4.8; R2 = 0.92/0.97)). Finally, the fields of application of both approaches are discussed. A major finding here is that the energy signature shows slightly better results at deriving the load profile when the measured heating demand is present from a previous heating season. Furthermore, the number of buildings does not affect the duration of a simulation because cumulative user profiles are used. Whereas the simulation duration of the physical approach depends essentially on the size of the investigated area. The physical approaches have the advantage of being able to include densification or renovation scenarios, demand forecasting and coupled simulations of buildings and smart energy systems of neighbourhoods.