Counting a few objects in one glimpse is easy. The ability to count (estimate number) is perfect in the "subitizing" domain of up to about 4 objects, but beyond that, error increases linearly with the number of target objects (Jevons, 1871; Kaufman et al., 1949). This is usually explained by supposing that a second, independent process provides estimates of many objects. We wondered whether crowding might affect one process more than the other, and thus help to distinguish them. Five observers performed several dot counting tasks. Visual conditions included surrounding the target dots, in a box, by distracter dots outside the box, using peripheral as opposed to central dot presentation, or both. We expected crowding only when the dots were peripheral and surrounded by distracters. Indeed, we find that this condition increases the standard deviation of the observers' counts by about 2 at every number of dots presented. This is trivial at large numbers, but dramatic at small numbers since the standard deviation is normally zero in the subitizing domain. Even so, the consistent increase of SD by +2 at all dot numbers seems difficult to reconcile with the popular two-process model of counting ability.