Microalbuminuria is known to predict mortality in elderly, non-insulin dependent diabetic individuals. To test whether this is also so when studied prospectively and over a longer period, we followed 228 persons with known diabetes mellitus and an age and sex matched, non-diabetic cohort for eight to nine years. Both cohorts were found during a population screening in 1981-1982 of 5292 citizens aged 60 to 74 years. At ascertainment extensive clinical and biochemical examinations including determination of urinary albumin were carried out. At May 15, 1990 111 diabetic and 46 non-diabetics had died (P less than 0.0001). In the diabetic cohort the median value of urinary albumin excretion (UAE) was 17.40 micrograms/min. In the group with values at or above 17.40 micrograms/min 62 died, compared with 40 deaths in the group with values below (P = 0.003). In the non-diabetic cohort the median UAE value was 7.52 micrograms/min. In the upper group 26 died, in the lower 15 (P = 0.05). Cox regression analyses showed coefficients of regression for ln(UAE) of 0.333 (P less than 0.001) for the diabetic group and 0.236 (P = 0.048) for the non-diabetic group. In the Cox model for the diabetics, ischemic heart disease was also of independent significance to mortality. The final model for the non-diabetics included hypertension and sex as significant variables. It is concluded that in a prospective study of elderly diabetics urinary albumin excretion rate is the best prognostic factor for long-term mortality.(ABSTRACT TRUNCATED AT 250 WORDS)