Four possible froms of bias in the evaluation of diagnostic information are evaluated. Reanalysis of a previously published experiment, in which radiologists evaluated excretory urograms, shows that the radiologists' assessment of the likelihood of a tumor being present was inappropriately influenced by the importance of ruling the tumor out. A new experiment is described in which fourth-year medical students and practicing pediatricians completed workups of hypothetical patients. In this study, both students and physicians subjectively distorted information in the latter part of the workup to support opinion formed up to that point. Also in evaluating diagnoses, students to a large degree and physicians to a lesser degree tended to ignore rates of disease prevalence. A fourth possible bias, that early, unreliable historical information dominates later, more reliable laboratory data, was not confirmed. Suggestions for overcoming the demonstrated biases, as well as other cognitive biases in the processing of diagnostic information, are made.