The geometry of the LOD1 block model of CityGML may be represented in a multitude of valid variants, e.g. it may represent individual buildings or aggregated blocks. The geometric representations within each of these, such as the reference of the footprint of a building, are not standardised in CityGML and are not informed in the metadata. The lack of knowledge of the used representation may possibly cause errors when the models are used for spatial analysis. Further, the effect of using different variants is not investigated. In this paper we (1) discuss this known, but frequently overlooked topic; (2) overview the possible geometric references and show how employing them may cause drastic differences for a GIS operation and/or use-case; (3) focus on the vertical (height) references for the top surface in LOD1, and show how to determine the optimal variant within a use-case with experiments using the Monte Carlo method; and (4) discuss the adoption and extension of metadata in INSPIRE to CityGML to reflect the employed geometric reference.