Determining compositions and functional capabilities of complex populations is often challenging, especially for sequencing technologies with short reads that do not uniquely identify organisms or genes. Long-read sequencing improves the resolution of these mixed communities, but adoption for this application has been limited due to concerns about throughput, cost and accuracy. The recently introduced PacBio Sequel System generates hundreds of thousands of long and highly accurate single-molecule reads per SMRT Cell. We investigated how the Sequel System might increase understanding of metagenomic communities. In the past, focus was largely on taxonomic classification with 16S rRNA sequencing. Recent expansion to WGS sequencing enables functional profiling as well, with the ultimate goal of complete genome assemblies. Here we compare the complex microbiomes in 5 cow rumen samples, for which Illumina WGS sequence data was also available. To maximize the PacBio single-molecule sequence accuracy, libraries of 2 to 3 kb were generated, allowing many polymerase passes per molecule. The resulting reads were filtered at predicted single-molecule accuracy levels up to 99.99%. Community compositions of the 5 samples were compared with Illumina WGS assemblies from the same set of samples, indicating rare organisms were often missed with Illumina. Assembly from PacBio CCS reads yielded a contig >100 kb in length with 6-fold coverage. Mapping of Illumina reads to the 101 kb contig verified the PacBio assembly and contig sequence. Scaffolding with reads from a PacBio unsheared library produced a complete genome of 2.4 Mb. These results illustrate ways in which long accurate reads benefit analysis of complex communities.