Each physician's ability to treat disease is limited by his/her ability to discriminate among patients on the basis of risk. The relationship of physician expectation of outcome (prognosis) to actual outcome and to cost were determined for 1,757 patients seeking primary care. Outcome was measured by a seven-level functional-status scale; patients who returned to their usual level of function after an acute illness were defined as having good outcomes. Although 24 per cent of patients experienced bad outcomes, physicians had anticipated only 6 per cent. Physician's predictions of bad outcomes had a sensitivity of 13.6 per cent and a specificity of 96.9 per cent. Patients with bad outcomes had slightly higher laboratory costs than did patients with good outcomes, but a much larger increase was seen in laboratory, office and total costs for all patients for whom physicians expected bad outcomes, regardless of the actual results. A feedback loop is recommended to provide a better perspective on outcome and eventually to improve the efficiency and cost benefit of the medical decision-making process.