Exposure to ionizing radiation is a known risk factor for breast cancer and the fertility pattern is a recognized modifier of breast cancer risk. The aim of this study was to elucidate the interaction between these 2 factors. This study is based on a Swedish cohort of 17 202 women who had been irradiated for skin haemangiomas in infancy between 1920 and 1965. The mean age at treatment was 6 months and the median breast dose was 0.05 Gy (range 0-35.8 Gy). Follow-up information on vital status, parity, age at first childbirth and breast cancer incidence was retrieved through record linkage with national population registers for the period 1958-1995. Analyses of excess relative risk (ERR) models were performed using Poisson regression methods. In this cohort, the fertility pattern differed from that in the Swedish population, with significantly fewer childbirths overall and before 25 years of age but more childbirth after that age. There were 307 breast cancers in the cohort and the standardized incidence ratio (SIR) was 1.22 (95% CI 1.09-1.36). A linear dose-response model with stratification for fertility pattern and menopausal status resulted in the best fit of the data. ERR/Gy was 0.33 (95% CI 0.17-0.53). In absolute terms this means an excess of 2.1 and 5.4 cases per Gy per 10(4) breast-years in the age groups 40-49 and 50-59 years respectively. The fertility pattern influenced the breast cancer risk in this irradiated population in a similar way to that observed in other studies. SIR at dose = 0 was highest, 2.31, among postmenopausal nulliparous women (95% CI 1.48-3.40, n = 62). SIR at dose = 0 was lowest in pre- or postmenopausal women with a first childbirth before 25 years of age; 0.89 (0.71-1.09) and 0.88 (0.58-1.25) respectively. Thus, in addition to the dose-effect response in the cohort, part of the breast cancer excess could be explained by a different fertility pattern. The estimates of ERR/Gy for the various categories of age at first childbirth, number of children, menopausal status and ovarian dose were very similar, contradicting any interaction effects on the scale of relative risk.