PacBio has a read length of more than 10kb, 50% longer than 5kb. 454 can now also up to 1kb.

http://en.wikipedia.org/wiki/DNA_sequencing#Next-generation_methods

I'm a little confused here -- you're doing amplicon sequencing right? Let's be clear too -- using marker gene primers is not metagenomics. You did not give us any information about your primer region -- why you are using that region? ...and what type of mixed sample you are looking to sequence?

I don't think PacBio would be the best use of your resources in regards to amplicon sequencing from an environmental sample -- you'll get the read depth you need with another technique, such as Illumina (more, shorter reads) or 454 (few expensive reads, but up to 1000 bp in length). I don't know why you would need to sequence a marker at 1500 bp in length. Why do you need something at that length? If you MUST use that specific region for your research question, you should be fine with flanking marker regions.

PacBio is still getting the kinks out with their technique, but it's mainly optimal for longer sequencing projects (i.e. whole genomes) especially where you need to close gaps from repeats and hard to assemble regions. The read depth is still an issue and the error rate would wreck havok on any amplicon sequencing project from an environmental sample. So in other words -- even if you were willing to pay the PacBio money -- it would not be the correct technology choice for your research question.

Syed,

It's true that PacBio is mostly used for genomic sequencing apps as Josh says, but recently we've been hard at work developing the platform for amplicon sequencing applications as well. Earlier this year I presented a poster on exactly this at ASM: http://files.pacb.com/pdf/Analysis_of_Full_Length_Metagenomic_16S_Genes_by_SMRT_Sequencing.pdf

Long story short, though our raw read accuracy requires some sophisticated algorithms to make the best use of, we can turn multiple passes over the same molecule into a high-quality "Circular-Consensus Sequence" which can mostly be analyzed with current tools. More over, with recent advances in sequence length we can now generate tens-of-thousands of reads of the length you desire from a single SMRTcell in ~2 hours. So if you're willing to put a bit more time and effort into the project on the analysis-side, I think our technology could suite your needs quite well.

Brett Bowman Scientist, Bioinformatics Pacific Biosciences, Inc.