In crowding, target perception is deteriorated by flanking elements. For example, when a vernier is flanked by two lines of the same length, vernier offset discrimination strongly deteriorates. Interestingly, changing the color of the flankers can reduce crowding (uncrowding). Similarly, when the flanking lines are part of a cube, i.e., a good Gestalt, crowding is weak. Conversely, however, scrambling the lines of the cubes, i.e., a “bad” Gestalt, leads to strong crowding. We proposed that crowding is strong when target and flankers group (two-lines, scrambled cubes). Crowding is weak when the target ungroups from the flankers (two flanking lines with different color, cubes). Here, we show, first, that when target and flankers group (strong crowding), crowding is unaffected by stimulus duration. Two same-length flankers presented for 20 ms lead to similar performance as when presented for 150 ms. Second, in uncrowding with flankers of different color (and other basic feature differences), performance is again unaffected by stimulus duration. Third, in uncrowding with cubes, duration matters. For short durations (20 ms), crowding is strong and only for longer stimulus durations (from 120 ms on) does crowding decrease. We suggest that, for short durations, the brain cannot process the good Gestalt of the cubes. The representation of the cube’s lines is “unstructured”, as in the scrambled cubes, and hence crowding is strong. Interestingly, a short preview (20 ms duration) of only the cubes strongly reduces crowding, even when the preview is presented one second before the “crowded” stimulus. Our results suggest that uncrowding emerges in a slow, recurrent manner, with iconic memory playing an important role.