Circular RNAs are thought to arise from non-canonical splicing of linear pre-mRNAs, as they frequently harbour the 3′ end of one exon joined to an upstream (rather than downstream) 5′ end of an exon. Zhang et al. used bioinformatic analyses on human transcriptome data and the functional testing of reporter constructs to dissect the sequence features underlying circularization. They found that exons joined during circularization are preferentially flanked by intronic inverted repeat elements (such as species-specific Alu repeats), which potentially cross-hybridize to bring these exon ends into proximity before splicing. In a separate study, Ashwal-Fluss et al. studied circular RNAs from flies and human cells. They found that RNA circularization occurs co-transcriptionally (as circular RNAs are also found in nascent transcriptomes). The second study also highlights the importance of flanking intronic sequences and additionally shows that RNA circularization may compete with canonical linear splicing. Both studies uncovered multiple alternatively circularized transcript isoforms for some genes, thus highlighting the complexity of RNA circularization.