MicroRNAs (miRNAs) are a group of small RNAs with regulatory roles at post-transcriptional level. Although they have been clustered based on their sequence or structure similarities, there is still no effective method to determine their functional similarities due to the lack of miRNAs functional annotation. To address this critical need, we presented here a novel method for systematic study of functional similarities among human miRNAs by using their target genes GO semantic similarities. The functional similarities were validated by comparing with miRNA expression similarities. To extract the highly significant clusters, we used multi-scale bootstrap re-sampling in clustering miRNAs functional similarities. The clustering of human miRNAs based on target genes molecular function annotation led to 44 significant clusters. The clustering results were coherent with biological knowledge. Our analysis suggests that systematic clustering based on target genes GO semantic similarities can aid to reveal the functional diversity of miRNA families. Additionally, this method can be extended to other species and used to predict novel miRNA functions.