Assembling individual genomes from complex community metagenomic data remains a challenging issue for environmental studies. We evaluated the quality of genome assemblies from community short read data (Illumina 100 bp pair-ended sequences) using datasets recovered from freshwater and soil microbial communities as well as in silico simulations. Our analyses revealed that the genome of a single genotype (or species) can be accurately assembled from a complex metagenome when it shows at least about 20 x coverage. At lower coverage, however, the derived assemblies contained a substantial fraction of non-target sequences (chimeras), which explains, at least in part, the higher number of hypothetical genes recovered in metagenomic relative to genomic projects. We also provide examples of how to detect intrapopulation structure in metagenomic datasets and estimate the type and frequency of errors in assembled genes and contigs from datasets of varied species complexity.