A constraint-based graphics system provides a flexible, intuitive framework for describing relationships among graphical objects in applications such as document preparation, font design, and solid modelling. This paper describes two constraint-based graphics systems, micro-COSM and the IDEAL Synthesizer, and their implementation in terms of attribute grammars. The implementation of these two systems is noteworthy since they represent the first interactive constraint-based graphics systems that are implemented using attribute grammars. Our experiences with attribute grammars suggest that they provide a powerful framework for representing constraints and extracting important semantic information such as the equations to be solved by the constraint solver. We discuss the advantages of using attribute grammars in constraint-based graphics and from our experiences make several observations about the way attribute grammars should be used.