By promoting of our government, more and more electronic business enterprise join the competition of online sales. With the sharp increase in sales, an ever increasing number of sales data is accumulated by the third party enterprise, and these sales data which is too scattered in original classification, it brings some difficulty in sales forecasting, in detail, it would lead to incomplete condition or severe deviation of predicted value. To improve this problem, a prediction model which is based on goods re-classification is constructed in this paper. This model used common sales features of products to extract the product cluster, then it used time series forecasting model to give the predicted value which is decorated by HMM in probability distribution aspect. Through experimental analysis, the final predicted values preferable fit the true values, and this achievement will provide the reference value to enterprise in establishing policies of distribution.