The industry trend towards self-service business intelligence is impeded by the absence, in commercially-available information systems, of automated identification of potential issues with summarization operations. Research on statistical databases and on data warehouses have both produced widely-accepted categorisations of measure attributes, the former based on general summarizability properties and the latter based on measures' additivity properties. We demonstrate that neither of these categorisations is an appropriate basis for precise identification of measure types since they are incomplete, ambiguous and insufficiently refined. Using a new categorisation of dimension types and multidimensional structures, we derive a measure categorisation which is a synthesis and a refinement of the two aforementioned categorisations. We give formal definitions for our summarizability types, based on the relational model of data, and then construct rules for correct summarization by using these definitions. We also give a method to detect whether a given MDX OLAP query conforms to those rules.