In this paper, we discuss a method for conceptual clustering of documents. Our cluster is defined with the notion of Formal Concept Analysis which can provide a conceptual meaning for each document cluster. Our clustering is formalized as a Top-N \delta-valid formal concept problem. We improve our previous clique search-based algorithm for the problem so that it can be applied to larger scale datasets. For more efficient computation, we present some pruning rules based on theoretical properties of formal concepts. A depth-first branch-and-bound algorithm with the prunings is designed. Our experimental results show valuable clusters can be extracted from a collection of web documents. Moreover, the algorithm outperforms some fast algorithms for mining closed itemsets equivalent to formal concepts.