To investigate what diagnostic strategies physicians adopt when the literature is unclear about the best use of diagnostic information, the authors examined how physicians weighted eight items of clinical information in diagnosing pulmonary embolism. Thirteen faculty members, 23 house officers, and 19 students estimated the likelihood of pulmonary embolism and predicted whether they would begin treatment in 27 simulated cases. The vignettes were constructed using a fractional factorial design and analyzed using conjoint analysis. Overall the lung scan results (34.7%) and leg examination (19.0%) were weighted most heavily in making the diagnosis, whereas the leg examination (35.7%) was given the greatest weight when deciding about therapy. Weights given by groups at different levels of training were similar, but there was considerable variation within groups. Heterogeneity of diagnostic strategies did not appear to decrease in groups with more training and experience. Multivariate analysis of predictors of pulmonary embolism in 102 actual cases showed that although lung scan results were important in both actual and simulated cases, heart rate accounted for the most variance in the actual cases but was hardly used in the physicians' strategies. There is considerable variation among physicians in how clinical information is used in diagnosing pulmonary embolism, and the variation may not decrease with increased experience.