This study examines whether the improvement in breast cancer survival in Hawaii suggested by an earlier analysis might be explained by concomitant temporal variations in prognostic factors. Characteristics at diagnosis and survival experience were compared for 1251 Caucasian, 1015 Japanese and 505 Hawaiian women diagnosed with invasive breast carcinoma during 1960-1979 and followed until the end of 1982. Time-trends were present for the patients' stage at diagnosis, age and socioeconomic status. A multivariate analysis using the Cox proportional hazards regression model revealed that adjustment for temporal variations in stage and age at diagnosis yielded a positive survival trend of a greater magnitude than that observed without adjustment, indicating that multivariate methods should be considered in time-trend analyses of cancer survival. Further analysis revealed that such improvement in breast cancer survival occurred for certain race-stage groups of patients only. Possible reasons for these trends in breast cancer prognostic factors and survival are discussed, along with potential biases. The data suggest that early detection might explain most of the survival improvement in Hawaii during the period of the study.