A normal scan for predicted microRNAs gives huge number of miRNAs in Target Scan. One strategy to pick the most probable candidate microRNA is to look for those microRNA binding sites which are highly conserved across vertebrates. You have this option in target scan. To reduce the numbers further, you can look in miRWALK, which gives a comparison of all the predicted microRNAs by different online tools. You can pick those microRNAs which are predicted by more 3 or 4 different online tools. This will ensure that you have picked those microRNAs which are consistent across different online tools as well as these are conserved across different species.

A normal scan for predicted microRNAs gives huge number of miRNAs in Target Scan. One strategy to pick the most probable candidate microRNA is to look for those microRNA binding sites which are highly conserved across vertebrates. You have this option in target scan. To reduce the numbers further, you can look in miRWALK, which gives a comparison of all the predicted microRNAs by different online tools. You can pick those microRNAs which are predicted by more 3 or 4 different online tools. This will ensure that you have picked those microRNAs which are consistent across different online tools as well as these are conserved across different species.

I would check one of the website with this specific databases such as mirnamap or similar (http://mirnamap.mbc.nctu.edu.tw/) which has a simple interface to look at.

I agree that miRWALK is very good: I've used it to determine which miRNAs to PCR-validate after a microarray. The prediction tool is useful to determine which miRNAs are more likely to target genes I'm interested in, and the validated targets tool allows me to get an idea of what is in the literature for either genes or miRNAs. Using validated targets, miRWALK will produce a list of pubmed references in which the gene of interest is associated with miRNAs. In this way one can get a overview over which miRNA-gene associations are backed up by the literature. There's quite a lot that's been published now, so depending on your research question this could be very useful, and easier than searching pubmed yourself. The tool does give a number of "false positive" associations, so you'll need to check the references carefully. The same can be done if (as in my case) a miRNA is your starting point and you want to know what biological pathways are associated to it in the literature

miRWalk is a collector for other algorithms while itself does not do much on the prediction. Based on such theory, it will summarize the prediction and providing the overlapping ones, however, it will also dilute the merits from each prediction algorithms for that it gives little or no priority concert for their assumptions. For example, miRScan considers 5 factors’ impacts with different weight (seed type, local context, location, 3' pairing, distance between sites, et al.. 2009 Bartel review paper, Cell) that do not affect the efficiency of the sites equally. PicTar has similar assumptions for this task. Several others emphasize the energy level on the base-pairing (miRBase) and others consider all match types equally.

In our hands, the only way you can find the “true” targets is screening and validation. The list of genes to be screened in reporter assays are coming from lists of genes generated as the cross-test from : a) expression levels inversely correlate with the miR; 2) miRScan or PicTar prediction with good scores; 3) conserved first, however, true target can also be non-conserved sites given the property of miR itself.