Fasting insulin has been used as a surrogate measure of insulin sensitivity in studies of non-insulin-dependent diabetes mellitus (NIDDM) risk, but the fasting insulin-NIDDM association may be confounded by insulin secretion, which correlates negatively with NIDDM risk and positively with fasting insulin level. In a prospective 5-year study of 137 nondiabetic Japanese-American men in King County, Washington State, higher fasting insulin was not strongly related to NIDDM (odds ratio (OR) = 1.37, 95% confidence interval (CI) 0.80-2.34), but this odds ratio increased substantially after adjustment for insulin secretion (OR = 2.92, 95% CI 1.41-6.06). Research on NIDDM risk in relation to fasting insulin may yield biased effect measures unless adjusted for insulin secretion.