Abstract: The algorithmic and evolutionary art movements within computer-generated art have helped spur interest in evaluating images on the basis of their aesthetic merit. When attempting to use non-interactive techniques to address this issue, two problems arise: (1) designing metrics that have explicit computational representations, and (2) establishing that such metrics actually fulfill their intended purpose. We survey our experiences in designing metrics for non-interactively guiding image evolution to obtain aesthetic images and we propose a taxonomy for metric frameworks. We also discuss some issues relevant to validating such metrics.