I am trying to generate 2 sets of 50 images, where each set is taken from a multivariate normal distribution with identity covariance matrix of size pxp, where p is the number of voxels in one of the images.

Is this possible using randn? I tried doing this by creating 100 all zero images, merging them with -t option into a 4D image 'mzeros' and then did

I then take images rand00...rand49 images to be one set, and images rand50..rand99 the other, which *appears* reasonable, but will this give me precisely a sample of the correct distribution for each set? I am unsure how the randn function works...

Cheers

Anil