To produce monthly employment and unemployment estimates for all 50 States, the District of Columbia and selected metropolitan areas, the Bureau of Labor Statistics (BLS) uses time series models applied to estimates from the Current Population Survey (CPS). The CPS design raises two types of problems for time series modeling. The first, and most obvious, is the variability in the data due to small samples. Secondly, the CPS has an overlapping design that induces strong autocorrelations in the survey errors (SE). When fitting time series models to CPS State data, it is important to explicitly account for the two important properties of the SE. This is done by using a signal plus-noise model where the monthly CPS estimates are treated as stochastically varying time series obscured by survey error. Given a model for the true labor force values (signal) and survey error (SE) variance-covariance information, we construct an estimator or filter that suppresses SE along with seasonal variation in the population. We also extend this model to a bivariate form to incorporate information in related series.