This paper takes a cognitive view of data modeling to address the usability of semantic modeling techniques and their potential impact on end user productivity. Specifically, we use a process tracing methodology to study the manner by which end users are able to use abstraction based data modeling to represent a reasonably complex problem and construct queries to address represented objects. The paper views the process of modeling as a constrained problem solving process. Drawing from research that has studied the systems design process, a cognitive model is developed and used to examine the semantic modeling process. We define a semantic modeling environment for the subjects of the experiment. Two specific abstractions are used: generalizations and composite objects. The representation constructs and the valid operations that are permitted on them are described. The results show that abstraction based modeling is a viable end user development methodology. Tentative recommendations for the design of platforms and training programs emphasize the need to encourage users to work at high levels of abstraction and utilize particular modeling heuristics that lead to better task performance.