Recent experiments have shown that the genomes of organisms such as worm, fly, human and mouse encode hundreds of microRNA genes. Many of these microRNAs are thought to regulate the translational expression of other genes by binding to partially complementary sites in messenger RNAs. Phenotypic and expression analysis suggest an important role of microRNAs during development. Therefore, it is of fundamental importance to identify microRNA targets. However, no experimental or computational high-throughput method for target site identification in animals has been published yet. Our main result is a new computational method which is designed to identify microRNA target sites. This method recovers with high specificity known microRNA target sites which previously have been defined experimentally. Based on these results, we present a simple model for the mechanism of microRNA target site recognition. Our model incorporates both kinetic and thermodynamic components of target recognition. When we applied our method to a set of 74 Drosophila melanogaster microRNAs, searching 3' UTR sequences of a predefined set of fly mRNAs for target sites which were evolutionary conserved between Drosophila melanogaster and Drosophila pseudoobscura, we found that a number of key developmental body patterning genes such as hairy and fushi-tarazu are likely to be translationally regulated by microRNAs.