There is conflicting evidence on whether the capacity of visual working memory (VWM) reflects a central capacity limit that also influences intelligence. We propose that encoding strategy and, more specifically, attentional selection, underlie the correlation of some VWM tasks and IQ, and not variations in VWM itself. In Experiment 1, change detection measures of VWM were found to be contaminated by some cognitive process that depressed performance at higher set sizes, so that fewer items were remembered when eight rather than just four were presented. Measuring VWM using whole report instead gave a less variable estimate that was higher, particularly for larger set sizes. Nonverbal IQ did not correlate with this estimate of VWM capacity, but instead with the additional factor that contaminates change detection estimates. We propose that this phenomenon reflects a lack of selection during encoding. In Experiment 2, we investigated the role of rehearsal using articulatory suppression and showed that this could not account for the key differences between the procedures.