Image analysis was used to investigate the prognostic significance of immunostaining for oestrogen receptor (ER), p53 tumour-suppressor protein and tumour cell proliferation (MIB-1) in a random cohort of 200 primary breast cancer patients with between 4 and 7 years of clinical follow-up. Image measurements of the percentage of immunopositive cancer cell nuclei (% positive nuclear area) were recorded for the above tumour features for each patient. Assessment of relative risk using Cox's univariate analysis indicated that tumour size, number of cancer-involved nodes, MIB-1 and ER % positive nuclear area were significantly associated with breast cancer disease outcome, i.e., relapse-free survival and overall survival. In multivariate analysis, tumour size, number of involved nodes, ER and MIB-1 % positive nuclear area were retained as independent predictors of prognosis, depending on the image measurement cut-point used. A prognostic model, which can be used without reference to nodal involvement, was constructed for tumour size, ER cut-point of 30% positive nuclear area and MIB-1 cut-point of 10% positive nuclear area. Kaplan-Meier analysis of this image-based prognostic index identified 4 risk groups with predicted 5-year overall survival rates of 93%, 83%, 76.7% and 61.5%. We conclude that image measurements of ER and proliferative rate can be combined with tumour size to construct a prognostic index which reliably predicts disease outcome in primary breast cancer without knowledge of the nodal status of the patient.