Many performance modeling systems are now graphically based, enabling the user to build a model by directly drawing and manipulating a pictorial model diagram on the computer display. These modeling systems generally provide a menu driven interface and an icon palette with which the user specifies the model by selecting and linbing icons and then providing associated textual attribute information. The resulting combination of graphics and text forms the model specification and requires that certain rules be obeyed in terms of content, completeness, syntax and semantics. Some classes of user errors can be completely prevented with this type of interface; others need to be detected and identified to the modeler, ideally as the model is constructed. Within the visual modeling paradigm, the detection and display of errors is significantly more complex than in text based modeling languages; the display problem is particularly acute in the case of multilevel, hierarchically structured models. In this paper, we describe the error detection and display capabilities we believe are needed in visual modeling tools and describe the implementation of these capabilities in the Research Queueing Package Modeling Environment (RES Q ME).