Traditional　subspace　clustering　algorithms　need　to　transform　each　sample　into　a　vector　form.　Therefore,　problems　of　high　dimensionality　and　small　size　samples　are　caused,　the　natural　structural　information　of　each　sample　is　ignored　and　the　clustering　information　is　missing.　To　overcome　the　drawbacks,　the　weighted　block　subspace　clustering　based　on　least　square　regression　algorithm　(WB-LSR)　is　proposed.　Firstly,　each　sample　is　divided　into　lots　of　blocks,　and　the　corresponding　affinity　matrices　of　each　block　are　obtained.　Next,　the　weight　of　each　affinity　matrix　is　determined　by　mutual　vote　between　affinity　matrices.　Finally,　the　weighted　sum　of　affinity　matrices　is　regarded　as　final　affinity　matrix.　The　experimental　results　on　image　datasets　and　motion　segmentation　video　datasets　show　that　the　proposed　method　effectively　improves　clustering　accuracy.　©　2016,　Science　Press.　All　right　reserved.