Text categorization is a very effective way to organize enormous number of documents in Digital Libraries. Accurate classification of documents is able to not only enhance document search precision, but also facilitate browsing-by-topic functionality. It is, nonetheless, difficult to obtain a satisfactory categorization accuracy compared to the corresponding results given by professional catalogers. This is due largely to the complexity of the pre-defined large-scaled category hierarchies that makes it difficult for learning algorithms to distinguish among categories. This paper describes a top-down document classification approach which takes advantage of the hierarchical structure, more specifically, in two ways: identifying the number of independent local classifiers and guiding top-down classification procedure. We finally evaluate it within the CINDI Digital Library applying ACM Classification System as targeted hierarchy. Experimental results show the promise of this approach.