MicroRNAs (miRNAs) are small non-coding RNAs that repress protein synthesis by binding to target messenger RNAs (mRNAs) in multicellular eukaryotes. The mechanism by which animal miRNAs specifically recognize their targets is not well understood. We recently developed a model for modeling the interaction between a miRNA and a target as a two-step hybridization reaction: nucleation at an accessible target site, followed by hybrid elongation to disrupt local target secondary structure and form the complete miRNA-target duplex. Nucleation potential and hybridization energy are two key energetic characteristics of the model. In this model, the role of target secondary structure on the efficacy of repression by miRNAs is considered, by employing the Sfold program to address the likelihood of a population of structures that co-exist in dynamic equilibrium for a specific mRNA molecule. This model can accurately account for the sensitivity to repression by let-7 of both published and rationally designed mutant forms of the Caenorhabditis elegans lin-41 3' UTR, and for the behavior of many other experimentally-tested miRNA-target interactions in C. elegans and Drosophila melanogaster. The model is particularly effective in accounting for certain false positive predictions obtained by other methods. In this study, we employed this model to analyze a set of miRNA-target interactions that were experimentally tested in mammalian models. These include targets for both mammalian miRNAs and viral miRNAs, and a viral target of a human miRNA. We found that our model can well account for both positive interactions and negative interactions. The model provides a unique explanation for the lack of function of a conserved seed site in the 3' UTR of the viral target, and predicts a strong interaction that cannot be predicted by conservation-based methods. Thus, the findings from this analysis and the previous analysis suggest that target structural accessibility is generally important for miRNA function in a broad class of eukaryotic systems. The model can be combined with other algorithms to improve the specificity of predictions by these algorithms. Because the model does not involve sequence conservation, it is readily applicable to target identification for microRNAs that lack conserved sites, non-conserved human miRNAs, and poorly conserved viral mRNAs. StarMir is a new Sfold application module developed for the implementation of the structure-based model, and is available through Sfold Web server at http://sfold.wadsworth.org.