This paper addresses the problem of automatically selecting the best among outputs from multiple machine translation (MT) systems. Existing approaches select the output assigned the highest score according to a target language model. In some cases, the existing approaches do not work well. This paper proposes two methods to improve performance. The rst method is based on a multiple comparison test and checks whether a score from language and translation models is signi cantly higher than the others. The second method is based on probability that a translation is not inferior to the others, which is predicted from the above scores. Experimental results show that the proposed methods achieve an improvement of 2 to 6% in performance.