Object-oriented simulation can represent objects and their behaviors and interactions in a cogent form that can be designed, evolved, and comprehended by domain experts as well as by systems analysts. Despite its many advantages, however, there remain several largely unexplored areas of need, requiring advances in the power and flexibility of modeling, in the representation of knowledge, in the integration of different modeling paradigms, and in the comprehensibility, scalability, and reusability of models. The Knowledge-Based Simulation project at RAND is working in several of these areas. This paper discusses the current limitations of object-oriented simulation, including modeling power, control representation, comprehensibility, and model building. The author also outlines some of the ways that paradigms can be extended to surmount these limitations.