In crowding, perception of a target is deteriorated by flanking elements. Crowding is usually explained by pooling models where target and flanker signals are averaged. We show here that crowding is rather determined by grouping and good Gestalt. We determined offset discrimination thresholds for verniers with different flanker configurations. When the vernier was flanked by two vertical lines, thresholds increased. Surprisingly, when the two lines were part of two cubes, thresholds decreased. This finding cannot be explained by pooling models, which predict stronger crowding for the cubes because more irrelevant lines are pooled. We explain our results in terms of grouping. When the target groups with the flankers, performance deteriorates (two-lines condition). When the flankers are part of a good Gestalt, the target ungroups from the flankers and performance improves (cube condition). For short durations (20-80 ms), thresholds were similarly high for the lines and the cubes conditions. For longer stimulus durations (160, 320 and 640 ms), thresholds stayed high for the two-lines conditions but decreased for the cubes condition. Our results show that Good Gestalt in crowding emerges “slowly” within 160 ms.