Abstract. Description Logic Programs (DLP) have been described as a description logic (DL) that is in the "expressive intersection" of DL and datalog. This is a very weak guideline for defining DLP in a way that can be claimed to be optimal or maximal in any sense. Moreover, other DL fragments such as EL and Horn-SHIQ have also been "expressed" using datalog. Is DLP just one out of many equal DLs in this "expressive intersection"? This paper attempts to clarify these issues by characterising DLP with various design principles that clearly distinguish it from other approaches. A consequent application of the introduced principles leads to the definition of a significantly larger variant of DLP which we conjecture to be maximal in a concrete sense. A preliminary report on the proof of this maximality is provided. While DLP is used as a concrete (and remarkably complex) example in this paper, we argue that similar approaches can be applied to find canonical d...