Importance of semantics in MLS database models

What is important when choosing the multilevel security model? Is it merely the way that model implements integrity rules (i.e. entity integrity property, referential integrity property) or should the model's semantics of what user on some specific level can see and/or believe be considered as well? In this paper, the basic concepts of multilevel security (MLS) database models are described and issues regarding the MLS semantics are considered. A comparison of well-known multilevel security models (LDV, Jajodia-Sandhu (1991), MLR, Smith-Winslett (1992) and Belief-Consistent Model) is given. The main problems, caused by the lack of semantics or incomplete semantic definitions, are discussed.