Recent　works　for　personalized　recommendation　typically　emphasize　their　efforts　on　learning　users＇　interests　from　interactions.　However,　users　make　decisions　depending　on　multiple　factors,　especially　various　attributes　of　items　like　appearance,　reviews,　price,　etc.　Therefore,　in　the　case　of　image　recommendation,　we　strive　to　unveil　users＇　interests　in　a　multimodal　manner.　In　this　work,　we　propose　a　multimodal　collaborative　graph　(MCG)　model　for　image　recommendation,　which　builds　users＇　interests　in　both　visual　and　collaborative　signals.　On　visual　modality,　visual　interest　filtering　is　designed　to　explore　the　interest　non-linearity　of　users＇　interacted　images.　In　the　pairwise　collaborative　module,　multi-hop　interactions　are　embedded　elaborately　to　encode　the　heterogeneous　structure　of　user-image　interactions　by　deep　interest　propagation.　Both　visual　and　collaborative　signals　are　aggregated　to　embed　users　and　items　and　match　pairwise　user-item　for　the　following　personalized　recommendation.　Experiments　are　conducted　on　three　public　real-world　datasets.　Further　analysis　demonstrates　the　compensation　capability　of　visual　and　collaborative　signals　in　mining　users＇　interests　and　verifies　the　effectiveness　of　the　proposed　MCG　for　image　recommendation.