While the potential prospects of domain-specific modelling languages (DSML) are undisputed, the design of a DSML faces specific challenges that have raised only little attention so far. They relate to the boundaries between a DSML and corresponding models as well as to the question how specific a DSML should be. Addressing these challenges does not only contribute to the development of meta modelling methods, it also relates to judging the economics of a DSML. This paper suggests guidelines to support typical decisions that occur with the design of a DSML. They mainly concern the level of abstraction a potential language concept should be defined on. The guidelines are not intended to serve as recipes for design, but rather to improve the transparency of design decisions.