We investigated whether model systems integrating stochastic variation into criteria for marker assessment could be used for monitoring metastatic breast cancer. A total of 3989 serum samples was obtained from 204 patients receiving first-line chemotherapy and from 112 of these patients during follow-up. Each sample was analyzed for cancer antigen 15.3, carcinoembryonic antigen, and tissue polypeptide antigen. The efficiency for identifying progression and nonprogression was 94% during therapy and 85% during follow-up, with no false-positive marker results for progressive disease. At clinical progressive disease, the median positive lead time was 35 days during therapy and 76 days during follow-up. Tumor marker assessment may document that a therapy is effective and ought to be continued in spite of adverse toxic effects, and that a treatment is ineffective and should be stopped to prevent unnecessary toxicity. Marker information may also be useful in studies investigating whether early treatment during follow-up will alter the prognosis of metastatic breast cancer.