Retrospectively cueing an item retained in visual working memory during maintenance is known to improve its retention. However, studies have provided conflicting results regarding the costs of such retro-cues for the noncued items, leading to different theories on the mechanisms behind visual working memory maintenance and retro-cueing. Here we tested an alternative explanation of the conflicting results regarding retro-cue costs—namely, that they are caused at least partly by differences in retro-cue reliability. We manipulated the ratio of valid-cue trials to invalid-cue trials within blocks. We used a continuous-report procedure that allowed fitting a model that provided recall probability and precision estimates for the memory representations. Reconciling previous contradictory findings, benefits for valid cues were observed in all conditions, but invalid cueing costs were found only when the retro-cue had a high reliability (i.e., was 80 % valid), but not when it had a lower reliability (i.e., 50 % valid). This was found for both the recall probability and the precision of visual working memory representations. Our results suggest that the cognitive mechanisms underlying retro-cue effects are strategically adjusted by participants, depending on the perceived retro-cue reliability.