miRRim is a method for detecting miRNA foldbacks based on hidden Markov model (HMM). In this method, the evolutionary and secondary structural features of a miRNA region is represented by a sequence of multidimensional vectors. HMMs that generate a sequence of continuous values are used to model the feature vector sequences. A miRNA model as well as three types of non-miRNA models are trained using feature vector sequences of the respective regions. These models are combined into a single HMM (Figure 1) and used to search genomic sequence for miRNA. The Viterbi decoding algorithm is used to determine a genomic segment which best fits the miRNA model, which corresponds to a predicted miRNA region.The stringency of miRNA prediction can be controlled by modifying the transition probability t from the non-miRNA models to the miRNA model. For more details, see Terai et al. (2007). 

Predicted results can be accessed at following links:
Human (hg18)
Fruit fly (dm3)

Figure 1 Hidden Markov model used to scan genomic regions. t is the transition probability from the non-miRNA models to the miRNA model.

Reference

miRRim: a novel system to find conserved miRNAs with high sensitivity and specificity.
Terai G, Komori T, Asai K, Kin T
RNA 13(12):2081-90, 2007 Dec, Epub 2007 Oct