There　should　be　different　requirements　for　False　Reject　rate　and　False　Accept　rate　in　classification　applications,　and　classifier　learning　should　use　an　asymmetric　factor　to　balance　between　False　Reject　rate　and　False　Accept　rate.　A　novel　AdaBoost　algorithm　was　developed　with　the　asymmetric　weight.　Moreover　we　provide　the　theoretical　analysis　of　its　performance　and　derive　the　upper　bound　of　the　classification　error.