Over three test sessions alternated with two training sessions, accuracy of 'same'/'different' judgments of four dry, white wines improved 14%. Confidence ratings were obtained so that the area under the isosensitivity curve could be computed as a bias-free measure of accuracy while the wine used for same pairs was held constant. An initial bias for 'different' judgments persisted, rather than disappearing as Walk found. While perceptual learning did occur, the problem of attending to irrelevant differences was not overcome.