In　recent　years,　with　the　rapid　growth　of　network　scale,　it　becomes　difficult　to　detect　communities　in　large-scale　networks　for　many　existing　algorithms.　In　this　paper,　a　novel　Parallel　Multi-Label　Propagation　Algorithm　(PMLPA)　is　proposed　to　detect　the　overlapping　communities　in　networks.　PMLPA　employs　a　new　label　updating　strategy　using　ankle-value　in　the　label　propagation　procedure　during　each　iteration.　The　new　algorithm　is　implemented　in　the　Spark　framework　for　its　power　in　distributed　parallel　computation.　Experiments　on　artificial　and　real　networks　show　that　PMLPA　is　effective　and　efficient　in　community　detection　in　large-scale　networks.