Four experiments were conducted to test whether conjunction errors were reduced after pictorial encoding and whether the semantic overlap between study and conjunction items would impact error rates. Across 4 experiments, compound words studied with a single-picture had lower conjunction error rates during a recognition test than those words studied with 2 pictures. This effect occurred even when participants were asked to respond quickly (Experiment 3) or to respond positively to any word that overlapped with the study phase (Experiment 2), as well as when the number of pictures shown at study was manipulated within-participants (Experiment 4). The effect of semantic overlap was ambiguous, with only Experiment 2 showing a difference between high and low overlap items. Overall, these results are inconsistent with a metacognitive interpretation of the role of pictures in reducing memory errors and are more consistent with impoverished familiarity after single-picture encoding.