This paper suggests that building models in a sorted first-order logic may be a useful inference mechanism in expert systems. Several kinds of inference that need to be performed in expert systems are presented, along with a discussion of the way a model-building approach applies to these tasks. However, the program that was built lacks the capability to work on all but the smallest size problems (in terms of number of clauses). If your interest is building expert systems to solve real problems today, this paper is not for you. This approach may eventually lead to better ways to build expert systems, but the research is still open.