Over the last 15 years, substantial progress has been made in understanding the potential and the limitations of the CA 125 assay. More than 2000 papers have been published concerning laboratory and clinical studies of CA 125. The original CA 125 assay utilized the OC 125 antibody that recognizes the CA 125 epitope on a high molecular weight glycoprotein. Despite repeated attempts, the gene encoding the peptide component has not yet been cloned. Monoclonal antibodies have been raised against other epitopes expressed by this molecule, leading to the development of the CA 125-II assay that exhibits less day-to-day variation. Using either assay, elevated levels of CA 125 are detected in a number of benign conditions, including endometriosis. CA 125 is most consistently elevated in epithelial ovarian cancer, but can be expressed in a number of gynecologic (endometrial, fallopian tube) and non-gynecologic (pancreatic, breast, colon and lung) cancers. The best established application of the CA 125 assay is in monitoring ovarian cancer. The rate of decline in CA 125 during primary chemotherapy has been an important independent prognostic factor in several multivariate analyses. Persistent elevation of CA 125 at the time of a second look surgical surveillance procedure predicts residual disease with > 95% specificity. Rising CA 125 values have preceded clinical detection of recurrent disease by at least 3 months in most, but not all studies. Given the modest activity of salvage chemotherapy, this information has not yet impacted on survival. Rising CA 125 during subsequent chemotherapy has been associated with progressive disease in more than 90% of cases. CA 125 may serve as an effective surrogate marker for clinical response in phase II trials of new drugs. CA 125 levels can aid in distinguishing malignant from benign pelvic masses, permitting effective triage of patients for primary surgery. Early detection of ovarian cancer remains the most promising application of CA 125. An algorithm has been developed that estimates the risk of ovarian cancer (ROC) based upon the level and trend of CA 125 values. A major trial has been initiated that uses the ROC algorithm to trigger transvaginal sonography and/or subsequent laparotomy. Such a trial could demonstrate improvement in survival through early detection. This strategy should provide adequate specificity, but sensitivity for early stage disease may not be optimal. In the future, improved sensitivity may be attained using multiple markers and neural network analysis. Most serum tumor markers have been proteins or carbohydrates, but lipid markers such as lysophosphatidic acid deserve evaluation. Genomic and proteonomic technologies should identify additional novel markers.