Shifting Bloom filters use location offset to encode state values for a set of elements. In spite of its novelty, classification error rate inherent to shifting Bloom filters needs to be improved. In this paper, we design a hierarchical shifting bloom filter to address this issue. Firstly, state values of elements in a set are partitioned into disjoint groups. Each group is assigned a unique group number. Then these group numbers implicitly assigned to elements are encoded by a major shifting Bloom filter(MShi ftBF). Each group is further associated with a secondary shifting Bloom filter(SShiftBF). State values of elements belonging to some group are encoded by the corresponding SShiftBF separately. Compared with standard Shifting Bloom filters, the advantage of the hierarchical structure of the proposed scheme is that we can improve both classification error rate and false positive rate. Finally, we provide theoretical analysis and conduct simulation experiments to demonstrate that the proposed scheme outperforms standard shifting Bloom filters.