This paper suggests a new semi-parametric multivariate approach to seasonal ad- justment. The primary innovation is to use a large dimensional factor model of cross section dependence to estimate the trend component in the seasonal decomposition of each time series. Because the trend component is speciÖed to capture covariation be- tween the time series, common changes in the level of the time series are accommodated in the trend, and not in the seasonal component, of the decomposition. The seasonal components are thus less prone to distortion resulting from severe business cycle áuc- tuations than univariate Ölter-based seasonal adjustment methods. We illustrate these points this using a dataset that spans the 2007-2009 recession in the US.