The course of the renal function and mortality were analysed in 395 patients with biopsy-proven glomerulonephritis (GN), using Cox's proportional hazards model. Seventeen clinical, biochemical and histopathological parameters were analysed for prognostic information. The patients were grouped according to their serum creatinine levels. Increase in serum creatinine, decrease in serum creatinine, cure and death were used as endpoints for the analysis. Caplan Meyer curves were made for 7 transitions between different groups and the variables were reduced by a step-wise procedure to a final model. Thirteen of the variables considered offered significant prognostic information (p less than 0.05) for at least one of the transitions. Short duration of disease, young age, non-nephritic urinary sediment and preceding streptococcal infection were predictors of cure. Extracapillary, membranoproliferative and unclassifiable GN, old age and arterial hypertension predicted increase in serum creatinine in patients with low serum creatinine, while male sex, short duration of disease and pathological electrocardiogram favoured a further increase in patients with high serum creatinine. A later decrease in serum creatinine was signified by a preceding streptococcal infection, short duration of disease, absence of arterial hypertension and low urinary protein excretion. Death without uremia was predicted by high age, connective tissue disease and extracapillary GN. Using these parameters and the models, it is possible to make a prognostic forecast for the individual GN patient. Examples of such a forecast are described.