Another approach for ICA estimation, inspired by information theory, is minimization of mutual information. We will explain this approach here, and show that it leads to the same principle of finding most nongaussian directions as was described above. In particular, this approach gives a rigorous justification for the heuristic principles used above.