Certain kinds of economic activity, and their associated time series, are affected significantly by holidays. When the date of a holiday changes from year to year, the effects of the holiday can impact two or more months in a way that depends on the date. In this case, the effect of the holiday is not confined to the seasonal component of the series. If such holiday effects are ignored, the models fit to the time series will often have reduced forecasting ability. Also the model residuals may show a lack of fit. Similarly, seasonal adjustments of the series may provide misleading signals in months affected by the holidays. In the U.S., the major moving holidays are Easter, Labor Day, and Thanksgiving, the dates of which vary over March 22-April 25, September 1-7, and November 22-28, respectively. Elsewhere, holidays tied to a lunar calendar, such as the Chinese New Year, Passover, and Ramadan have an economic impact. Although we only apply the modeling and model selection procedures described in this paper to model U.S. holidays, they are applicable more generally.