Hi Meg, 

It looks like you have lots separate gradient direction files (with a low number of directions). Seeing that you use denoising and Gaussian smoothing and still have bad fits, I suspect that your gradient directions might be too dependent on one another for accurate fitting. The best way to test this is to look at the condition number of the matrix used for the weighted linear least squares ( if this matrix is ill-conditioned, things will go wrong). I understand that this is not a straightforward thing to do without the source code. However, we are working on that!

Just to make sure. Can you send an example image with "bad voxels" that you are trying to get rid of?  

I hope this helps,
Emilie