With　the　rapid　development　of　social　networking,　user　requirement　suffers　more　and　more　from　intention　gap　of　user　interest　and　semantic　gap　of　multimedia.　It　becomes　urgent　to　investigate　personalized　recommendation.　In　this　paper,　we　propose　modular　manifold　ranking　(MMR)　for　image　recommendation.　MMR　attempts　to　construct　global　image　manifold　over　CNN　based　features　extraction　to　involve　content　relations　in　recommendation.　Specifically,　manifold　modularity　is　introduced　to　perform　a　flexible　manifold　learning　for　large　scale　database　in　a　manner　of　manifold　decomposition.　MMR　employed　manifold　ranking　to　propagate　users＇　interests　to　the　whole　image　manifold　and　estimate　user-image　correlation　for　image　recommendation.　The　experimental　analysis　illustrates　that　the　proposed　method　successfully　extends　the　scalability　of　manifold　ranking　and　achieves　good　performance　in　social　image　recommendation　compared　with　its　competitors.