ABSTRACTType II CRISPR-Cas9 systems require a small RNA called the trans-activating CRISPR RNA (tracrRNA) in order to function. The prediction of these non-coding RNAs in prokaryotic genomes is challenging because they have dissimilar structures, having short stems (3–6 bp) and non-canonical base-pairs e.g. G-A. Much of the tracrRNA is involved in base-pairing interactions with the CRISPR RNA, or itself, or in RNA-protein interactions with Cas9. Here we develop a new bioinformatic tool to predict tracrRNAs. On an experimentally verified test set the algorithm achieved a high sensitivity and specificity, and a low false discovery rate (FDR) on genome analysis. Analysis of representative RefSeq genomes (5462) detected 275 tracrRNAs from 165 genera. These tracrRNAs could be grouped into 15 clusters which were used to build covariance models. These clusters included Streptococci and Staphylococci tracrRNAs from the CRISPR-Cas9 systems which are currently used for gene editing. Compensating base changes observed in the models were consistent with the experimental structures of single guide RNAs (sgRNAs). Other clusters, for which there are not yet structures available, were predicted to form novel tracrRNA folds. These clusters included a large and divergent tracrRNA set from Bacteroidetes. These computational models contribute to the understanding of CRISPR-Cas biology, and will assist in the design of further engineered CRISPR-Cas9 systems. The tracrRNA prediction software is available through a galaxy web server.