The medical decision making literature has previously considered the test and test-treatment thresholds in a normative fashion. In the normative approach, the analyst calculates the optimal threshold--the likelihood of disease at which testing or treatment should be undertaken. In contrast, we describe a method of deriving the threshold in a descriptive fashion, by determining the probabilities of disease at which clinicians actually make the decision to test or to initiate specific treatment without further testing. In applying this method, the analyst first asks clinicians to provide an estimate of the prior probability of disease, and to select one of three options: test, treat, or do neither. After receiving new information about the patient, the clinicians are asked to revise the probability estimate and to select a new option. Correlation of changes in the probability of a disease with changes in the clinicians' selections of options to test or treat enables the analyst to estimate the test and test-treatment thresholds used by the clinicians in medical decision making. Knowledge of these thresholds also enables the analyst to calculate the clinicians' ratio of the benefits to the costs of the therapy being considered, considering the risks of the test itself.