We determined the degree of variability and sampling distribution of several commonly used parameters of microalbuminuria in patients with non-insulin-dependent diabetes mellitus (NIDDM) and proposed a sampling strategy for estimating the level of albuminuria. Four patients with NIDDM with previously documented microalbuminuria collected 30 consecutive split (overnight and daytime) 24-hour urine samples (experiment A). These samples were analyzed for total 24-hour albumin excretion; daytime, overnight, and 24-hour albumin concentration; and daytime, overnight, and 24-hour albumin-to-creatinine ratio. In a second experiment (B), 10 patients collected 10 consecutive overnight urine samples. Finally, a total of 300 separate triplicate urine samples were analyzed for the variability of 24-hour albumin excretion (100 samples) and albumin-to-creatinine ratios in 24-hour urine (100 samples) and overnight urine (100 samples). We found that the sampling distribution shape of all parameters of albuminuria is positively skewed, without consistent evidence of log-normality. When two methods were used for quantifying day-to-day variability (the interquartile range/median ratio and the chance of a single measurement being >50% off the actual value of albuminuria), the overnight albumin-to-creatinine ratio is the least-variable parameter of albuminuria, scoring 0.38% and 10% on both methods, respectively, in experiment A. Collecting multiple samples of overnight urine improves accuracy. The largest gain in precision in estimating the actual value of albuminuria is obtained for sample sizes of three and five and does not increase with nonconsecutive sampling of urine. Based on the combined data from experiments A and B, the expected mean deviation of the median of three and five overnight samples from the actual level of the overnight albumin-creatinine ratio is 17.9% and 12.1%, respectively. An analysis of variability in three sets of 100 triplicate 24-hour urine samples shows that the overnight albumin-to-creatinine ratio is a significantly more-constant parameter of microalbuminuria than the amount of albumin excreted in 24 hours or the albumin-to-creatinine ratio in 24-hour urine (p < 0.05). We concluded that the parameters of diabetic albuminuria have positively skewed, non-log-normal sampling distributions. The overnight albumin-to-creatinine ratio is the least-variable parameter of microalbuminuria. We recommend collecting three consecutive early morning urine samples, using the median value of the albumin-to-creatinine ratio in these samples for quantifying albuminuria.