Both　massive　multiple-input　and　multiple-output　(MIMO)　and　physical-layer　network　coding　(PNC)　techniques　can　effectively　improve　the　network　throughput.　However,　for　PNC-aided　massive　MIMO　systems,　the　high-order　modulations　superimposed　at　the　relay　particularly　bring　computational　burden　and　performance　degradation　to　the　signal　detection.　Thus,　this　letter　proposes　an　iterative　MIMO-PNC　detector　from　double　expectation　propagation　(DEP)　as　a　high-accuracy　solution　in　this　large-scale　system.　By　exploiting　the　PNC　characteristic,　the　summation　and　difference　(SD)　information　of　the　user　pair　messages　are　first　linearly　formed　and　iteratively　updated　between　the　DEP　detector　and　the　channel　decoder.　Simulation　results　show　that　the　proposed　iterative　method　can　achieve　significant　performance　gains　of　2　dB　in　various　massive　MIMO　systems,　as　compared　to　the　non-iterative　one　and　traditional　linear　minimum　mean　square　error　(LMMSE)　detector.