Are subitizing and counting implemented as separate or functionally overlapping processes?

Enumeration of small groups of four or fewer objects is very fast and accurate (often called "subitizing"), but gets slower and more error prone for more than four items ("counting"). Many theories have been proposed to account for this dichotomy, most suggesting that "subitizing" and "counting" are two qualitatively different and separable processes. Others, in contrast, have proposed that the two operations reflect two different levels along a continuum of complexity. In this paper we present a PET study that attempts to characterize subitizing and counting at a neural level in order to investigate whether they are implemented as separate or functionally overlapping processes. Subjects performed an enumeration task on visual arrays of dots that varied in numerosity (1-4 and 6-9 dots) and spatial arrangement (canonical and random). The results demonstrated a common network for subitizing and counting that comprises extrastriate middle occipital and intraparietal areas. The intensity and spatial extent of this network were modulated by the number of dots and their spatial arrangement: activation increased as the number of items in the visual array increased, reaching maximum peak and extent for counting 6-9 randomly arranged items. Direct comparison between subitizing and counting showed that counting, relative to subitizing, was correlated with increased activity in this occipitoparietal network, while subitizing did not show areas of increased activation with respect to counting. Results speak against the idea of the two processes being implemented in separable neural systems.