Dear Tor, Doug and others, we did some fMRI measurements and simulations on the use of various ISI distributions for detecting the main effect (its in Neuroimage (14:1193,2001) if you want to have a look). Indeed, we found that the exponential distribution was the best, both for detecting the response and estimating the shape of the hemodynamic response (HDR). These characteristics stems from the high variance associated with it that leads to high t-values and hence a high detection power (see Friston et al.,NeuroImage 12:196 2000) and also the FIR-model yields high values for the "shape-estimator" (Dale, Human Brain Mapping 8:109,1999). To put it in pictures: short ISI's lets the BOLD "build up" to high levels (i.e. high detection power) while pauses in the event train lets the BOLD come back to baseline again ( and hence the shape can be estimated). These slight empirical findings, based on simulations, only holds for a linear time invariant system, and an HDR modelled by two gamma-functions (the one SPM99 uses for the regressors). For a FIR-model it may be different as Doug points out. In the experimental setting, we also found high activations with other ISI timings. cheers Gisela -- Gisela E. Hagberg, PhD Functional Neuroimaging Laboratory Foundation Santa Lucia Via Ardeatina 306 I-00179 Rome Voice: +39.06.5150.1349 FAX: +39.06.5150.1213 E-mail : [log in to unmask]