We discuss recent progress towards an empirical test of ‘realism’ in cognition, ‘realism’ in this context being the property that cognitive variables always have well defined (if unknown) values at all times. Our main result is an inequality obeyed by realist theories, which could be tested by a suitable experiment. We focus our attention in this contribution on two particular issues. The first is the exact notion of realism which is to be tested, as this has received less attention in earlier work. The second is an important technical issue about the inequality we use; in earlier work Atmanspacher and Filk (2010) considered a different expression, and we explain why our inequality is more suitable for use under realistic experimental conditions.