36 patients with insulin-dependent diabetes mellitus who had 'Albustix'-negative urine but raised urinary albumin excretion (30 to 300 mg/24 h) were randomly assigned to either remaining on conventional insulin treatment or continuous subcutaneous insulin infusion and followed up for 2 years. The insulin-infusion group showed a significant, sustained improvement in metabolic control, with a median glycosylated haemoglobin of 7.2% (range 5.9-8.8), but there was no change in the conventional-treatment group (median 8.6%, range 7.2-13.4) (p less than 0.001). Clinical diabetic nephropathy (a urinary albumin excretion rate above 300 mg/24 h in at least two of three 24 h urine collections) developed in 5 patients in the conventional-treatment group, but not in the insulin-infusion group (p less than 0.05, two-tailed). Fractional albumin clearance (mean and range X 10(7] increased in the conventional-treatment group from 160 (35-468) to 360 (29-1580) and was unchanged in the insulin-infusion group (170 [31-608] before to 160 [26-460] after) (p less than 0.05). Insulin infusion had an overall beneficial effect on the annual increase in urinary albumin excretion (p less than 0.05), and the mean glycosylated haemoglobin values correlated positively with annual change in albumin excretion (r = 0.57, p less than 0.0001). The diastolic blood pressure rose significantly in the conventional-treatment group (p less than 0.001), and annual change in mean blood pressure correlated with change in urinary albumin excretion (r = 0.49, p less than 0.001).