Michael,

Once you fed in 3dDeconvolve with a set of lags, you would get what you asked for with the modeled impulse response function (output of iresp), which would not directly tell you how good the fit is.

If you initially assume that the hemodynamic response from a stimulus lasts for 12 s and TR = 2 s, you run 3dDeconvolve and get results with minlag and maxlag of 0 and 6 respectively. There is no single and direct quantity to judge the goodness of the fit, but you can check the following few things: full F statistic, R^2, the comparison between the orignal signal and the fitted time series, activation patterns, etc..

If you can't run 3dDeconvolve due to some intrinsic design issue of the experiment (non-invertible error), you would have to try with different set of lags.

Gang