Abstract: Forecast is critical to sales companies, and its accuracy will directly affect profits and long-term development. Considering the unique life cycle features of cosmetics sales, the combination forecasting model based on CBR (Case-Based Reasoning) is proposed. The CBR can compensate for insufficient data of short life cycle products, and reasoning and calculating the targets' attributes according to the "most similar" are not only more accurate, but also easy to practice. On this basis, the paper puts foreword the improved BASS model and seasonal trends growth model, giving full consideration to seasonal and trend fluctuations in the data series, as well as its complete life cycle characteristics. And then, the AFTER-GA (Aggregated Forecast Through Exponential Re-weighting-Genetic Algorithm) is proposed, taking PSOWv-SVM for its ability to deal with external factors, highlighting the strengths of each model, and so to obtain higher accuracy. Example shows that the AFTER-GA is superior to other methods in accuracy; meanwhile, the combination forecasting model based on CBR can achieve a satisfactory accuracy, and has practical value for cosmetics sales forecast.