Bundling is an important strategy for online supermarkets to promote sales. A proper bundle can not only bring profits to the retailer, but also can provide convenience to consumers. However, the number of products sold by the online supermarkets are extremely large, and customers may vary at different cities. How to generate bundles from a large number of products and according to the consumption characters of different cities is a major challenge for online supermarkets. In this paper, a product bundling model considering different city levels was proposed. First, the frequent itemset were found based on the analysis of association rules. The sales contribution rate was introduced to identify the profitability of frequent itemset for different city levels, and based on which an optimizing model was formulated to select proper itemset as bundles. The results of numerical experiments show that bundles can be effectively selected across different city levels, and by bundling the income of the seller can be increased.