To generate hypotheses about how physicians make difficult clinical decisions, we analyzed transcripts of the "thinking aloud" behavior of expert clinicians making a testing or treatment decision with an uncertain diagnosis. We compared the clinicians' reasoning with a decision analysis of the same problem. The experts did not formulate a global outline of their decision, but chained together a sequence of decisions based on available and incomplete information. Despite effective and efficient problem solving, the clinicians used numeric terms only as symbolic representations of likelihood, used limited information in choosing among alternatives, and dismissed the possibility that a less conventional strategy, empiric therapy, might yield equivalent outcome. We describe cognitive problem-solving strategies and knowledge representations that permit persons to make successful decisions despite limited processing resources. The same cognitive procedures probably contribute to observed errors in decision-making under uncertainty.