This paper examines the time series properties of state and national unemployment rates. Based upon unit root, variance ratio, and cointegration tests, as well as Granger-causality and error-correction model results, several important conclusions can be made. First, forecasting models that include only levels of unemployment rates may produce spurious regression results. Second, in the vast majority of cases, there is no long run co-movement between the aggregate US unemployment rate and individual state unemployment rates. Third, models that are specified in first-differences generally yield reliable insights into state-national unemployment relationships.