A four-year prospective study of the factors predicting albuminuria was carried out in 172 normotensive, insulin-dependent diabetic patients without overt nephropathy. Urinary albumin excretion was estimated as the urinary albumin:creatinine ratio (UA/UC) in an early morning sample. Multivariate analysis showed that UA/UC on the return visit was positively associated with the UA/UC (p less than 0.001) and glycosylated haemoglobin (HbA1; p less than 0.001) at initial examination; weaker associations were found with a history of hospital admission (p less than 0.05) and smoking (p less than 0.05), and with treatment of blood pressure (p less than 0.05). Neither initial blood pressure, heart rate, nor creatinine clearance were significant predictors of the UA/UC. Two patients died from coronary heart disease, both of whom had raised albumin excretion at initial examination. Eleven (6.8 per cent) of the 160 patients who were studied repeatedly developed macroalbuminuria (UA/UC greater than 45.5 mg/mmol): they had a significantly higher initial UA/UC (p less than 0.005), HbA1 (p less than 0.05) and a greater frequency of retinopathy (p less than 0.05) than patients matched for age, sex and duration of diabetes who did not develop macroalbuminuria. Simultaneous measurements of the UA/UC and HbA1 should be used when screening for microalbuminuria in diabetes mellitus: patients with a high UA/UC (e.g. greater than 3.5 mg/mmol) and HbA1 (e.g. greater than 13 per cent) should be closely monitored even when blood pressure is normal.