Pigeons were trained in a conditional discrimination paradigm to diffe
rentiate successively presented visual arrays according to the relativ
e number of their elements. Transfer tests with novel stimuli demonstr
ated that they discriminated the categories of ''many'' (6 or 7) from
''few'' (1 or 2) items. In further tests, other new stimuli were intro
duced that consisted not only of these training numerosities, but also
of the intervening ones (3, 4, and 5). Variations in the birds' discr
imination performance corresponded to the order of stimuli on a numero
sity dimension. This serial ordering was maintained when other factors
such as brightness, size, shape, area, and contour of the elements we
re systematically controlled across tests. Smaller numerosities were s
omewhat better discriminated than those at the higher end of this test
range.