A major limitation in conducting functional neuroimaging studies, part
icularly for cognitive experiments, has been the use of blocked task p
aradigms. Here we explored whether selective averaging techniques simi
lar to those applied in event-related potential (ERP) experiments coul
d be used to demonstrate functional magnetic resonance imaging (fMRI)
responses to rapidly intermixed trials. In the first two experiments,
we observed that for 1-sec trials of full-field visual checkerboard st
imulation, the fMRI blood oxygenation level-dependent (BOLD) signal su
mmated in a roughly linear fashion across successive trials even at ve
ry short (2 sec and 5 sec) intertrial intervals, although subtle depar
tures from linearity were observed. In experiments 3 and 4, we observe
d that it is possible to obtain robust activation maps for rapidly pre
sented randomly mixed trial types (left-and right-hemifield visual che
ckerboard stimulation) spaced as little as 2 sec apart. Taken collecti
vely, these results suggest that selective averaging may enable fMRI e
xperimental designs identical to those used in typical behavioral and
ERP studies. The ability to analyze closely spaced single-trial, or ev
ent-related, signals provides for a class of experiments which cannot
be conducted using blocked designs. Trial types can be randomly interm
ixed, and selective averaging based upon trial type and/or subject per
formance is possible. (C) 1997 Wiley-Liss, Inc.