To organize the large number of products listed in a large e-commerce product catalog, each product is usually assigned to one of the many categories in the multi-level taxonomy tree. It is a time-consuming and difficult task for merchants to select proper categories within thousands of options for the products they sell. We'll explain how we built an automatic classification tool for predicting the correct category for a given product title and description. We used a combination of two different neural models -- deep belief nets and deep autoencoders -- to categorize products using both titles and descriptions. We'll also show how we scaled out with GPUs to categorize a large e-commerce dataset consisting of 172 million products and a 5-level taxonomy tree containing 28,338 categories.