Hi, > Another possibility is to decompose it into rotate, zoom, rotate with > singular value decomposition, and then reconstruct without the zooms (as > used by Procrustes analysis): > > [U,S,V]=svd(M(1:3,1:3)); > R=U*V' Just in case it's helpful (it was to me), this thing above is polar decomposition, that gives the closest pure rotation matrix from a matrix that may contain shears (as well as zooms). http://en.wikipedia.org/wiki/Polar_decomposition Best, Matthew