Mind-wandering (MW) often involves a decoupling between attention and external information (Schooler et al., 2011). The present study examined whether eye movements during MW decouple from image content in a scene perception task. Participants studied real-world scenes and occasionally answered thought probes assessing their attentional states (on-task, intentional MW, and unintentional MW). We built salience maps (Graph-Based Visual salience; Harel, Koch, & Perona, 2007) and meaning maps (Henderson & Hayes, 2017) to represent how low- and high-level image features were distributed across the scene. Meaning values at viewed locations showed that participants prioritized meaningful regions over less meaningful ones even during MW. But a subsequent analysis incorporating unexamined locations showed that participants tended to overlook meaningful regions during MW compared to when being on-task. Results for image salience were similar to the results for meaningfulness, as meaning and salience were highly correlated (r = 0.84). When directly compared to each other, meaning maps outperformed salience maps in predicting fixations regardless of participants’ attentional state. In addition, there were no significant differences between intentional and unintentional MW in all analyses. In sum: (1) MW affected visual attention differently at the fixation level and the trial level: individual fixations preferred meaningful regions even during MW, but overall, attention was insufficient to cover meaningful regions in the scene. (2) Relatedly, meaning maps were less precise at predicting fixation allocation during MW compared to when being on-task. (3) Even during MW, eye movements appeared to be driven more by meaning than by salience.