A new algorithm for the diagnostic evaluation of erythrocytosis is presented. This algorithm is based on principles of decision-making analysis and on the pathophysiology of erythrocytosis. The initial task is to identify smokers, because of the high probability that they have "smoker's" polycythemia. The current diagnostic criteria for polycythemia vera may be insufficient when the probability of disease is low but may be too rigorous when the probability of disease is high.