The ordinal discrete labels are usually obtained from continuous labels, and the se regres sor seldom use the mutual membership information between ordinal discret e labels, which can be further improved . Therefore, quantitive representation is character ized for the membership information, and then a kernel discriminant learning for ordinal regression using label membershi p (LM KDLOR) is established by combining the representation with typical off t he shelf KDLOR. Experimental results with the standard ordinal regression data sets verify the e ffectiveness of the proposed strategy.