We used a pipeline comprising several packages to annotate identified TFs. Domain structure and GO annotation were predicted by InterProScan (version 4.8) (23 (link)). Cross-links to well-known resources were assigned to the best BLAST hits with maximal e-value 1e-10. Nuclear localization signals were predicted by PredictNLS (24 (link)). Other information such as expert-curated description, expression, regulation, conserved elements and references was collected from corresponding databases. Multiple sequence alignments (MSAs) for DBDs were constructed by HMM-guided method, and MSAs for full-length protein sequences were inferred by T-coffee (version 9.03) (25 (link)). Family trees across 83 species were inferred by FastTree (version 2.1.3) (26 (link)) with 100 resamplings. Family trees within each species were inferred by MrBayes (version 3.2.1) (27 (link)) based on the Dayhoff model for 50 000 generations. The Help page (http://planttfdb.cbi.pku.edu.cn/help_info.php#tfinfo) describes more detailed information on datasets and parameter settings.