Proper categorisation of light vehicles is crucial for analysing and comprehending the developments taking place in the road transport sector, that impact the environment, road safety, transport operation, and urban planning. However, current vehicle classification methods in Europe are based on empirical or legacy approaches, sometimes founded on obsolete criteria, and do not fully reflect recent changes in the vehicle fleet and market. This paper aims to establish a scientific approach for the classification of light vehicles by introducing a Bayesian statistical method to define vehicle segments in an explicit and reproducible way. Contrarily to previous studies that mostly depend on machine learning techniques, which, despite their high accuracy typically lack explainability, the proposed approach prioritises the transparency of classification decisions. Through an in-depth examination of vehicle physical attributes, key variables were identified and utilised to determine clear boundaries between segments. These boundaries were articulated through simple linear relationships of the chosen variables, thus providing well-defined criteria open for interpretation and verification. The algorithm could assign up to 82\% of the vehicles to the original segments. The accuracy demonstrated is comparable to that of several unsupervised machine learning models and transparently reveals the boundaries among different segments. The findings can be used by researchers and modellers to update existing vehicle fleet models, particularly those treating environmental and energy consumption impacts, or used as a possible standard for multi-purpose vehicle classification.