Urban building energy models (UBEMs) are expected to play a key role in the integrated assessment of sustainability measures on both district and city level. However, due to limited availability of data sources, those models are often created through an archetype approach, which is a deterministic method to allocate building envelope characteristics to building groups. Unfortunately, this deterministic approach may underestimate the variability of the existing building stock, which is important when designing district energy systems to optimise the location of production and storage units within the system. In contrast to the deterministic approach, this work presents a new probabilistic approach to allocate building envelope characteristics within UBEMs that in combination with stochastic occupants enables to include the variability of existing districts. A thorough comparison of the deterministic and the probabilistic method is established for 820 buildings of the Boxbergheide district in Genk by performing dynamic energy simulations in the IDEAS Modelica library. For the studied district, a probabilistic building envelope characterisation with standard occupants increases the coefficient of variation (CV) on the energy demand for space heating, compared to a deterministic approach with standard occupants, from 17.8% to 46.4%. Including a probabilistic building envelope characterisation increases the variability on the energy demand for space heating to a larger extent than including stochastic occupants, which increases the CV to only 29.6%.