Background: We introduce DESMAN for De novo Extraction of Strains from MetAgeNomes. Metagenome sequencing generates short reads from throughout the genomes of a microbial community. Increasingly large, multi-sample metagenomes, stratified in space and time are being generated from communities with thousands of species. Repeats result in fragmentary co-assemblies with potentially millions of contigs. Contigs can be binned into metagenome assembled genomes (MAGs) but strain level variation will remain. DESMAN identifies variants on core genes, then uses co-occurrence across samples to link variants into strain sequences and abundance profiles. These strain profiles are then searched for on non-core genes to determine the accessory genes present in each strain. Results: We validated DESMAN on a synthetic twenty genome community with 64 samples. We could resolve the five E. coli strains present with 99.58% accuracy across core gene variable sites and their gene complement with 95.7% accuracy. Similarly, on real fecal metagenomes from the 2011 E. coli (STEC) O104:H4 outbreak, the outbreak strain was reconstructed with 99.8% core sequence accuracy. Application to an anaerobic digester metagenome time series reveals that strain level variation is endemic with 16 out of 26 MAGs (61.5%) examined exhibiting two strains. In almost all cases the strain proportions were not statistically different between replicate reactors, suggesting intra-species niche partitioning. The only exception being when the two strains had almost identical gene complement and, hence, functional capability. Conclusions: DESMAN will provide a provide a powerful tool for de novo resolution of fine-scale variation in microbial communities. It is available as open source software from https://github.com/chrisquince/DESMAN.