Various filtering algorithms for publish/subscribe systems have been proposed. One distinguishing characteristic is their internal representation of Boolean subscriptions: They either require conversions into DNFs (canonical approaches) or are directly exploited in event filtering (non-canonical approaches). In this paper, we present a detailed analysis and comparison of the memory requirements of canonical and non-canonical filtering algorithms. This includes a theoretical analysis of space usages as well as a verification of our theoretical results by an evaluation of a practical implementation. This practical analysis also considers time (filter) efficiency, which is the other important quality measure of filtering algorithms. By correlating the results of space and time efficiency, we conclude when to use non-canonical and canonical approaches.