Do you mean that you put a whole row of 0's in the glt matrix? It that is the case, it would definitely make 3dDeconvolve unhappy about it because such a row of 0's can't be implemented in general linear test.
If you really want to keep the original meaningful matrix intact, I fail to see why you could not make
that row with 1's in appropriate places instead of all 0's. In doing so, you could run 3dDeconvolve as usual. In the end, it would not hurt anything if you set up with those 1's in
that row because general linear tests have
NO effect whatsoever on the deconvolution/regression analysis. Instead glt only generates the test results the user requested after the deconvolution/regression analysis is done. If you don't hope to analyze that specific test, you can pretend you did not run such a test and don't look at the results in AFNI although they are stored in a few subbriks. :)
Gang