In previous work, we used a comprehensive sliding-window molecular clock analysis of near full length HIV-1 genomes to find that different regions of the virus genome yield significantly different estimates of the time to the most recent common ancestor. This finding, together with other evidence of the deep recombinant history of SIV and HIV, is not consistent with the standard model of the evolutionary history of HIV-1/M where non-recombinant subtypes are globally distributed through founder effects, followed by occasional inter-subtype recombination. We propose to re-examine the history of HIV-1/M using recombination detection methods that do not rely on predefined non-recombinant genomes. Here we describe an unsupervised non-parametric clustering approach to this problem by adapting a community detection method developed for the analysis of dynamic social networks. We compared our method to other reference-free recombination detection programs, namely GARD (HyPhy) and RDP (versions 4 and 5). We simulated recombination events by swapping randomly sampled branches in a time-scaled tree, which we reconstructed from subtype reference sequences in BEAST. Extant sequences were simulated for each tree using INDELible, and concatenated segments delimited by randomly sampled breakpoints from the resulting alignments to form the recombinant sequences. We show that our community detection method outperforms GARD, RDP4 and RDP5 in detecting recombinant breakpoints in simulated data, with a significantly lower mean error rate (Wilcoxon test, P < 0:05). Our method groups HIV-1 into 25 communities and detects evidence of inter-subtype recombination in pure subtype reference genomes obtained from Los Alamos HIV sequence database. For instance, we estimate that sub-subtypes A1 and A2 may contain large fragments from subtype C, while subtype C seems to contain fragments from subtype G. Our method provides a new reference-free framework for detecting recombination in viral genomes, and network communities may provide an alternative framework for HIV-1 classification.