Our perception of the visual world is rich and full of detail, but our ability to retain this information, even over brief periods of time, is limited. In this study, we investigated whether observers allocate these precious limited resources optimally. Three colored disks were briefly presented and then disappeared. Following a short delay, one disk reappeared with either the same or a different color, and subjects reported whether or not a change occurred. Two types of trials were tested that differed in the fidelity of encoding necessary to perform the task. In High Discriminability (HD) trials, the changes were large, requiring only low fidelity encoding of the colors. In Low Discriminability (LD) trials, the changes were smaller, and as expected, subjects performed significantly worse on the more difficult LD trials. Is this performance difference due to a fixed resolution of visual memory, or to a difference in encoding strategy between the conditions? To address this question, we next asked subjects to perform the task in mixed trial blocks biased in the number of HD and LD trials with an 80/20 ratio. Interestingly, a significant improvement in performance was observed on LD trials when tested in blocks with a greater number of HD trials. Thus, it appears that subjects were employing a suboptimal encoding strategy for blocks with a greater number of LD trials. These results indicate that there is some flexibility in memory resource allocation and that under some circumstances, observers underestimate their capacity and resort to suboptimal encoding strategies.