This paper examines the relationship between area-level social capital and non-specific psychological distress. It demonstrates that not controlling for non-time-varying omitted variables can seriously bias research findings. We use data from three cross-sections of the US National Health Interview Survey (1999, 2000, and 2001): 37,172 observations nested within 58 Metropolitan Statistical Areas. We also add data from the Area Resource File and County Business Patterns. We use a validated measure of social capital, the Petris Social Capital Index (PSCI), which measures structural social capital. We estimate a two-level multilevel linear model with a random intercept. Non-specific psychological distress is measured using a valid and reliable indicator, the K6. Individual-level variables include sex, age, race/ethnicity, marital status, education, family income, smoking status, exercise status, and number of visits to a health professional. Area-level covariates include the PSCI, the unemployment rate, psychiatrists per 1000 population, non-psychiatric physicians per 1000 population, and area-level indicators to account for non-time-varying area-level omitted variable bias. Time dummies are also included. We find that lagged area-level social capital is negatively related to non-specific psychological distress among individuals whose family income is less than the median. These associations are much larger when we control for non-time-varying area-level omitted variables.