RNA plays a crucial role in post-transcriptional regulation. Similar to transcriptional regulation, post-transcriptional regulation is often accomplished by the binding of proteins to specific motifs in mRNA molecules. Unlike DNA binding proteins, which recognize motifs composed of conserved sequences, RNA protein binding sites are more conserved in structures than in sequences. A lot of works have been done for RNA structure prediction; however, most of them focus on single RNA structure prediction instead of finding characteristic structure motifs within a RNA family. Though some current approaches can now identify common structure motifs from a set of RNAs, they typically assume the given set forms a single family, which is not necessarily correct. We propose a new adaptive method that conducts structure prediction and clustering simultaneously. Its performance is demonstrated on several real RNA families.