If the spatial normalization has too much freedom in the distortions that it can estimate, then there is more chance of things going wrong. What you could do is either reduce the number of nonlinear basis functions that you are using (or restrict the registration to an affine transformation) or try increasing the amount of regularization by typing something like: global sptl_Rglrztn sptl_Rglrztn = 0.001; before running the spatial normalization. For SPM96, decreasing the value of the global variable sptl_Rglrztn will mean that less variability is allowed in the deformations. Something else that I might suggest is that you create your own template image so that the contrast in the template image is the same as that in the images you wish to normalize. This can improve matters considerably. Regards, -John > in some of our data we have observed the same phenomenon > as discribed earlier in the spm newslist by Martin Rausch > and later by Justin O'Brien this year. > > http://www.mailbase.ac.uk/lists/spm/1998-07/0125.html > http://www.mailbase.ac.uk/lists/spm/1999-01/0108.html > > Unfortunately we do not find answers to this question. Has > anyone had the same trouble and if yes what do you do about it? %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%