Existing model-based knowledge-acquisition tools can acquire large knowledge bases and update these knowledge bases as knowledge changes. These tools, however, are brittle. They can only be used to acquire knowledge for a particular problem solver performing a specific task and they are not easily adapted to new problem solvers. Brittleness limits the effectiveness of these tools because the dynamic nature of knowledge systems make modifications both necessary and frequent. This paper presents a model of knowledge systems that reduces brittleness by separating acquisition techniques for search-control knowledge from other types of knowledge, by driving knowledge acquisition from properties of a knowledge-level description of the task instead of the problem solver, and by using ontologies to reuse knowledge bases.