A microRNA (miRNA) is a small noncoding RNA molecule about 22 nucleotides in length. The paper describes a web server for predicting miRNAs and their precursors and binding sites. The predictions are based on either sequence similarity to known miRNAs of 223 organisms or context-structural hidden Markov models. It has been shown that the proposed methods of prediction of human miRNAs and pre-miRNAs outperform the existing ones in accuracy. The average deviation of predicted 5'-ends of human miRNAs from actual positions is 3.13 nt in the case of predicting one pair of complementary miRNAs (miRNA-miRNA* duplex). A useful option for our application is the prediction of an additional miRNA pair. In this mode, the pairs closest to actual miRNA deviate by 1.61 nt on average. The proposed method also shows good performance in predicting mouse miRNAs. Binding sites for miRNAs are predicted by two known approaches based on complementarity and thermodynamic stability of the miRNA-mRNA duplex and on a new approach, which takes into account miRNAs competition for the site. The role of the secondary structure in miRNA processing is considered. The web server is available at http://wwwmgs.bionet.nsc.ru/mgs/programs/rnaanalys/.