It is now clearly established that proteolytic enzymes, including plasminogen activator (uPA), play an important role in breaking down the extracellular matrix, which is considered to be a step in metastasis formation. Plasminogen activators are controlled at various levels. Two inhibitors, PAI-1 and PAI-2, have been identified, the latter being more specific for uPA. In attempts to determine their prognostic value, it is essential to investigate the relative importance of these parameters and their interactions. We used an immunoenzymatic method to assay uPA, PAI-1 and PAI-2 antigens in cytosols prepared from 314 primary breast tumours. The patients were followed up for a minimum of 6 years and all relevant clinical and laboratory findings were recorded. Univariate analysis confirmed the poor outcome of patients whose tumours contained large amounts of uPA and PAI-1. In addition, low levels of PAI-2 correlated with shorter disease-free survival in the overall population (P = 0.02), post-menopausal women (P = 0.02) and women without lymph node involvement (P = 0.02). Multivariate analysis in the 'main effects' Cox model identified node involvement, macroscopic tumour size and PAI-2 as significant variables. The 'interactive' model, taking into account interactions between uPA and its two inhibitors, identified a first subgroup with a very poor prognosis associating either high levels of PAI-1 with low levels of PAI-2 in the overall population and the women with no node involvement or high levels of uPA with low levels of PAI-2 in the group of menopausal women. We conclude that PAI-1 provides the same prognostic information as uPA, and does not appear to play a role as an inhibitor. In contrast, PAI-2 increases the prognostic value of uPA, particularly in post-menopausal women, and PAI-1 in patients with no node involvement.