Evaluating the effectiveness of virtual environments as training and analysis systems one must take into account both strongly and weakly defined measures of visual behaviour and associated experience. The investigation of cross correlations between strongly defined measures of logged gaze behaviours, and weakly defined measures of subjective perceptions of visual behaviour, reveals significant discrepancies. The existence of these discrepancies casts doubt upon the effectiveness of using self-reporting questionnaires to assess training effectiveness. However, making participants aware of these discrepancies can be a potentially powerful method for increasing the effectiveness of training using virtual worlds.