A retrospective, population-based study was undertaken to determine variations in the management of women aged less than 50 years with primary breast cancer in different hospital settings and the influence of these variations on survival. A total of 1757 women who were resident in the South East Thames Health Region aged less than 50 years at the time of diagnosis of breast cancer and who presented during a 5 year period (January 1984 to December 1988) were recorded by the Thames Cancer Registry. The hospitals at which primary surgery was undertaken were categorised as teaching or non-teaching hospitals. The non-teaching hospitals were grouped according to the mean number of patients treated annually during the study period (< or = 2, 3-9, > or = 10 each year). The following factors were compared between these groups: age, extent of disease, tumour morphology, extent of primary surgery (mastectomy vs less than mastectomy), use of axillary surgery (any vs none) and use of systemic adjuvant therapy. Survival rates for the different groups were compared. Registration rates did not differ significantly between health districts. A total of 1485 (85%) women underwent surgery in over 90 different hospitals. In 1324 (86%) of these cases the surgery was undertaken in a total of 42 NHS hospitals within SE Thames Health Region or in seven teaching hospitals in adjacent regions. Mastectomy rates decreased from 52% in 1984 to 28% in 1988 (P<0.0001), but were consistently higher in teaching hospitals (P=0.01). The use of any form of axillary surgery decreased from 49% to 36% over the 5 year period (P=0.003), with significantly lower rates of axillary surgery being performed in non-teaching hospitals (P<0.0001). The proportion of cases recorded as having non-specific morphology was higher in nonteaching than in teaching hospitals (P<0.0001). On multivariate analysis survival was significantly (P<0.001) influenced by stage and tumour histology. Among patients who underwent surgery, the type of hospital in which this was undertaken did not appear to influence survival significantly. This analysis of routine cancer registry data indicates that patients were widely dispersed in a large number of different hospitals and that there were marked variations in practice according to the type of hospital to which patients presented. The treatments provided were frequently at variance with those recommended at a consensus conference held during the study period, particularly in relation to the use of axillary surgery and adjuvant systemic therapy. The way in which services are currently provided may hamper the delivery of appropriate management and comprehensive support. These data thus have implications for the purchasing and provision of services for this common condition.