The objectives were to measure roundfish selectivity and to test if square-mesh windows inserted in the codend could improve the selectivity. Sea trials were conducted with a commercial trawler in the Skagerak area. Three codend types were tested: (1) a standard codend with 104 mm meshes; (2) a standard 104 mm codend with two 85 rum square-mesh side windows; (3) a standard 104 mm codend with an 85-mm square-mesh top window. The twin-trawl method was used where one side of the rig had a 35-mm (nominal mesh size) control codend. Hauls of each codend were fitted simultaneously in a fixed- and random-effects model. In total, 24 and 30 hauls of haddock and cod, respectively, were included in the final selectivity analysis. This indicated that insertion of a top window in a standard codend improved the selectivity resulting in a lower selectivity ratio (SR/L50) and a higher L50 for cod and haddock, whereas insertion of side windows gave a lower selectivity ratio for both species, but a higher L50 for cod only. The results suggest that the recent mesh size increase to 120 mm in a standard codend is not sufficient to reduce the probability of discards to a negligible level with the existing minimum landing sizes. (C) 2004 Elsevier B.V. All rights reserved.