On each trial a target object and a fragment of the target (or a control st
imulus) were presented briefly enough to be integrated together. The stimul
i were masked, and identification accuracy was measured. The fragments were
large or small in size scale, and were presented early in processing (frag
ment before target) or late in processing (fragment after target). When pre
sented early, large-scale fragments tended to facilitate identification mor
e than small-scale fragments, but when presented late, small-scale fragment
s facilitated more than large-scale fragments. Facilitation effects from co
mmon feature fragments supported the idea of a spatiotemporal dependency, i
n which the efficiency of processing a piece of information depends on othe
r pieces of information that have been processed. This is a strong type of
global-to-local processing and can be interpreted within a structural descr
iption framework.