There are various things to consider: 1. To determine microbiome composition using 16S rRNA directly will tell you the active fraction, not the composition of the community. To do so, you need to extract total RNA, normally rRNA is around 90% of it, and then to cDNA and dsDNA by Reverse Transcription + PCR. 2. If you mean amplifying 16S rRNBA genes using primers annealing in conserved sites from metagenomic (environmental) DNA), then the next point 3. would be if the bacterial biomass per ml is enough to have an idea of the microbiome composition or not. i.e. if you have 10 exp +5 cells per ml and you have a DNA extraction of >90% efficiency, then you are just in the threshold (near a total of 1 ng) of what could be amplifiable as PCR template, and to determine the community structure. 4. Depends of the technique, but if you are planning to use next generation sequencing techniques in that amplicon, and you expect to obtain >10.000 reads per amplicon, then is better to be always above number cells extracted  vs reads expected. 5. Then, if you have in your sample an extract with a representative amount of total metagenomic DNA, then the point is how to extract it. One is to filter. But if the amount is less than 100 ml, just use centrifugation at maximum speed to pellet all cells. If you have less than 5 ml, you can use it to a lysis buffer and do standard bacterial DNA extraction with it. i.e http://femsle.oxfordjournals.org/content/272/2/269.full If you have very low amounts of total DNA extract, one option is to apply phi29 isothermal amplification to obtain enough template (I did it for some specific applications requiring it http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0052069), but if you start with a very low amount of metagenomic DNA, then is not so representative. And this is rather expensive.