According to the long term and seasonal trends of time series, time-series decomposition makes a reasonable forecast of future, but when dealing with seasonal factors in China, it’ll be influenced by Chinese traditional festivals. Based on time-series decomposition, this paper built a modified model consisting of time-series decomposition and dummy variables which represent Chinese traditional festivals. In an example of the 90 months’ cigarette sales forecast in a province, the new model can effectively improve the prediction accuracy, and it’s helpful for enterprises to make production and sales plans.