Rfam is a curated database of non-coding RNA families, each represented by multiple sequence alignments, consensus secondary structures and covariance models (CMs). Our families may be divided into three broad functional classes: non-coding RNA genes, structured cis-regulatory elements and self-splicing RNAs. Rfam uses covariance models which simultaneously model RNA sequence and structure to provide better discrimination than purely sequence based models. Rfam families are created from European Nucleotide Archive sequence data and experimental evidence in the literature. In addition to alignments and CMs, we provide ontology terms and external references, as well as search tools to enable users to query their sequence against Rfam data. Rfam is used for automatic annotation of genome sequences as well as a test dataset for many RNA bioinformatics methods.