Motivation: MicroRNAs (miRNAs) are small non-coding RNAs that are extensively involved in gene expression regulation. One major roadblock in functional miRNA studies is the reliable prediction of genes targeted by miRNAs, as rules defining miRNA target recognition have not been well-established to date. Availability of high-throughput experimental data from a recent CLASH (cross linking, ligation and sequencing of hybrids) study has presented an unprecedented opportunity to characterize miRNA target recognition patterns, which may provide guidance for improved miRNA target prediction. Results: The CLASH data were analysed to identify distinctive sequence features that characterize canonical and non-canonical miRNA target types. Most miRNA targets were of non-canonical type, i.e. without involving perfect pairing to canonical miRNA seed region. Different miRNAs have distinct targeting patterns, and this miRNA-to-miRNA variability was associated with seed sequence composition. Specifically, seed-based canonical target recognition was dependent on the GC content of the miRNA seed. For miRNAs with low GC content of the seed region, non-canonical targeting was the dominant mechanism for target recognition. In contrast to canonical targeting, non-canonical targeting did not lead to significant target downregulation at either the RNA or protein level.