I have presented a review and critique of the procedures employed in simultaneous discrimination training experiments using octopuses as subjects. Procedural variables were analyzed statistically for their influence on experimental outcome. The variables most significantly associated with successful discriminations included use of a specific start location for subjects, shock as negative reinforcement, fewer trials per session, more sessions per day, and discriminations based on stimulus brightness. No experiment controlled all potential sources of inadvertent cues, and subjects' performances appeared to be sensitive to exact procedural details. The most common practice diminishing evidence for learning involved reward that coincided with the subject's pre-existing preferences. I found no evidence that sub-optimal experimental designs biased experimental outcomes in any significant and systematic way. Although there is sufficient reason for rejecting results of published simultaneous discrimination training experiments, careful conclusive experiments remain to be performed.