Would you please tell us what you did to create your regressors and tell us which regressors are different between the two models?
Regressors are based on the onsets times, durations upsampled to the number of time bins convolved with the HRF from spm_get_bf. They are then downsampled to the TR. Regressors are not demeaned, or orthogonalized. Derivatives are orthogonal based on the basis functions being orthogonal. Parametric modulators are demeaned before forming the PM regressor.
Stimulus time -- if in the time bin the stimulus is present, its coded as 1. This vector of stimulus times is convolved with the HRF.
In SPM, the data is also scaled to a global mean of 100.
There are also potentially filters and autoregressive terms as well.