Pham, depending on what sort of datasets you have (Illumina?, 454 pyro?) always work with OTUs from denoised/chimera checked datasets.  You will need then to assign OTUs to a taxonomic rank usually genus or up (species is more problematic usually due to resolution limitations) and then just calculate proportions of the OTUs within the rank you want. If you get a group of OTUs not classifiable at your favored rank you can only really assign these to a higher rank (potentially as high as phylum-level). You should have no or very few unassigned OTUs. If you cannot assign any I would be suspicious and I would BLAST these sequences to assess quality and their cellular association (mitochondrial, chloroplast etc.). I found for Illumina and pyro data class level works best but this depends on the samples your are working with and the relative diversity of groups. For GI tract work genus/family is great; for soil - class level is better

 If you have Illumina data just use the 16S app on the basespace cloud server it will save a lot of time to the point this classification process becomes a trivial exercise,

Good luck with your research 

Dear  John P Bowman,

Thank you very much for your suggestion.