This paper presents the results of comparing three control strategies for expert systems: event driven, expectation driven, and goal driven. Three different versions of a pulmonary function analysis system (PUFF) were implemented, each with a different control strategy. The systems are described and compared for efficiency and naturalness. The advantages and disadvantages of each strategy are discussed. The reasons why one approach, the expectation-drive strategy, is best suited for the PUFF application are summarized.