In order to study the relationship between circulating levels of CA 15-3 and the disease extent in predicting survival, we prospectively followed 312 breast cancer (BC) patients, from October 1988 to March 1995, from the time of first relapse. CA 15-3 values were assessed before treatment onset. Disease extent was defined as the percentage of liver or lung involvement and the number of bone segments positive at scintigraphy. The covariates were primary tumour characteristics (T, N and hormone receptor status) and patient characteristics at recurrence (menopause, performance status and age). Higher CA 15-3 serum levels were found in patients with visceral metastases or with pleural effusion. A logistic regression model selected disease extent in liver, lung and bone as independent variables for the determination of abnormal CA 15-3 values. Univariate survival analysis confirmed the positive prognostic influence of low CA 15-3 serum levels, absence of visceral metastases and the presence of only one metastatic site. Multivariate Cox's survival analysis selected disease extent in liver, lung, bone and soft tissue but not level of CA 15-3 as prognostic factors. In conclusion, CA 15-3 is not an independent variable in determining survival, its prognostic role being linked to the disease extent. This association suggests that CA 15-3 may be useful in assessing disease extent when this is not easily assessable.