Edge　detection　has　been　widely　used　in　medical　image　processing,　automatic　diagnosis,　et　al.　A　novel　edge　detection　algorithm,　based　on　the　fusion　model,　is　proposed　by　combination　with　the　two　proposed　models　as　follows:　the　matrix　of　most　probable　distribution　of　edge　point　and　the　matrix　of　the　difference　weight　of　each　point.　The　most　probable　distribution　of　edge　point　can　be　obtained　by　analyzing　the　variance　among　4-connected　neighborhood　points　around　each　pixel　under　estimation　in　the　image　to　label　the　all　candidate　edge　points　in　the　image.　The　difference　weight　of　each　point　can　be　gotten　by　analyzing　the　brightness　difference　between　the　neighborhood　point　and　the　under-estimating　pixel　to　represent　the　probability　of　being　edge.　The　two　matrices　gotten　from　the　different　descriptions　of　spatial　structure　are　fused　together　and　derive　from　the　final　edge　image　with　thresholding　method　on　the　fusion　matrix.　The　experiments　are　performed　based　on　the　public　diabetic　retinopathy　database　DRIVE.　According　to　the　edge　images　obtained,　the　proposed　method　is　subjectively　analyzed　to　be　complete　and　close　to　the　Ground　Truth　image　with　very　low　noise　in　comparison　with　the　Sobel,　Canny　and　LOG　edge　detectors.　The　F1　measure,　ROC　measure　and　PFOM　measure　are　separately　adopted　to　make　quantitative　evaluation　of　the　proposed　edge　detection　algorithm.　Experimental　results　show　that　the　proposed　method　is　able　to　improve　the　effect　of　edge　detection　on　medical　images.