The　classification　and　measurement　of　rock　fragments　is　very　important　in　mining　and　construction　engineering.　The　monitoring　system　acquires　and　analyses　the　fragment　images　from　a　gravitational　falling　stream　at　the　end　of　a　moving　conveyor　belt,　and　the　key　function　of　the　system　is　to　construct　an　image　segmentation　algorithm.　To　achieve　this　goal,　an　adaptive　thresholding　algorithm　with　fuzzy　comprehensive　evaluation　is　proposed.　Firstly,　a　grabbed　image　is　roughly　segmented　by　using　a　global　auto-thresholding　algorithm.　Then　each　of　the　objects　is　measured　and　analyzed　if　it　includes　the　multiple　fragments　touching　each　other,　based　on　the　fuzzy　comprehensive　evaluation　method　in　which　the　salient　fragment　features　of　area,　perimeter,　shape,　gradient　magnitude　and　gray-level　flatness　are　extracted,　and　for　each　of　the　features,　the　membership　function　is　constructed　experimentally.　Finally,　each　of　the　touching　fragment　regions,　as　one　image,　is　auto-thresholded　again,　and　this　procedure　is　repeated　until　no　region　can　be　further　separated.　The　experimental　results　show　that　compared　to　cluster　analysis,　graph　based,　and　FCM　image　segmentation　algorithms　the　new　algorithm　can　make　image　segmentation　well　for　the　falling　fragments　on-line.　©　2015　SERSC.