In comparative fishing method, one of important experimental methods in research on size selectivity of trawls, the control codend mesh size is crucial and a possible upper limit is generally suggested to be half the mesh size of the test codend. However, the suggestion is unavailable when the body size of target specie of trawl is small enough, especially in shrimp or prawn trawl fisheries. In order to discuss how to choose a rational mesh size of control codend, experiments, in which test codends with different mesh sizes (20, 30, 35 and 40 mm) were rigged in a same multicodends beam trawl and fished simultaneously, were carried out in Lvsi Fishing Ground. The catch frequency of Parapenaeposis hardwickii, Pseudosciaena polyactis and Collichthys lucida retained by codends were fitted to a selectivity analysis model in which the selectivity curves of different codends were regarded as geometrical similar under two contrary hypotheses (hypothesis 1: the 20 mm mesh size codend is regarded as test codend and selective for catch species and, hypotheses 2: as control codend and nonselective for species). The result demonstrated that the goodness of model fit and the selective parameters of different test codends were not affected significantly by the two contrary hypotheses. Taking the difference in relative fishing intensity and statistical method of analyzing size selection between multiconends trawl and conventional trawls into account, authors suggest that the assumption of nonselectivity of 20 mm mesh size control codend is acceptable in analysis of size selectivity in the experiemts, which means that the general upper limit of control codend mesh size could be broken.