Multi-view　clustering　has　attractive　intensive　attention　and　proved　to　be　more　effective　than　single-view　clustering.　The　mining　and　effective　utilization　of　information　complementarity　is　core　of　multi-view　clustering.　In　this　paper,　we　propose　a　novel　Auto-Weighted　Graph　Regularization　and　Residual　Compensation　for　Multi-view　Subspace　Clustering　(ARMSC)　method,　which　adopts　residual　representation　information　to　compensate　the　globally　low　rank　consensus　representation.　The　auto-weighted　multi-view　graph　regularization　term　is　constructed　to　preserve　the　local　manifold　structure　of　multiple　features,　which　can　automatically　adjust　the　weights　and　the　learned　weights　are　adopt　to　measure　the　importance　of　each　view＇s　residual　representation　for　final　representation.　Specifically,　we　formulate　the　graph　regularized　consistent　representation　using　nuclear　norm　and　a　set　of　group　effect　residual　compensation　using　Frobenius　norm.　Finally,　we　introduce　a　convex　relaxation　and　alternating　direction　method　to　optimize　the　problem.　Comprehensive　and　ablation　experiments　on　seven　real-world　data　sets　illustrate　the　effectiveness　and　superiority　of　the　proposed　ARMSC　over　several　state-of-the-art　multi-view　clustering　approaches.