It　is　quite　difficult　to　register　two　pathological　retinal　images　owing　to　the　vague　vascular　network　existence.　Moreover,　general　registration　methods　based　on　detecting　the　crossovers,　bifurcations　of　the　vascular　network　have　certain　drawbacks.　In　this　paper,　a　retinal　image　registration　method　based　on　invariant　feature　is　proposed.　We　recognize　the　extracted　invariant　feature　as　feature　point　and　proposed　bilateral　＂or＂　the　best-bin-first　(BBF)　algorithm　to　match　feature　point.　In　addition,　we　use　the　feature　points＇　orientations　and　the　geometrical　property　to　exclude　the,　error　matches　and　utilize　the　refined　matching　features　pair　to　estimate　transformation　by　M-estimation　for　retinal　image.　The　registration　results　are　observed,　the　correct　matches　are　compared　and　the　root　of　mean　square　error　is　analyzed　by　experiments　of　registering　various　degrees　of　pathological　retinal　images.　Results　show　that　the　proposed　method　achieves　fine　detail　alignment,　reserves　sufficient　correct　matches　and　the　root　of　mean　square　error　is　less　than　1　and　around　0.5　for　pathological　retinal　images　of　registration　success.　Experimental　results　validate　the　accuracy　and　effectiveness　of　the　proposed　method.