Introduction: Selective attention to threat-related information has been associated with clinical delusions in schizophrenia and nonclinical delusional ideation in healthy individuals. However, it is unclear whether biased attention for threat reflects early engagement effects on selective attention, or later difficulties in disengaging attention from perceived threat. The present study examined which of these processes operate in nonclinical delusion-prone individuals. Methods: A total of 100 psychologically healthy participants completed the Peters et al. (1999) Delusions Inventory (PDI). Twenty-two scoring in the upper quartile (high-PDI group) and 22 scoring in the lower quartile (low-PDI group) completed a modified dot-probe task. Participants detected dot-probes appearing 200, 500, or 1250 ms after an angry-neutral face pair or a happy-neutral face pair. Results: High-PDI individuals responded faster to dot-probes presented in the same location as angry compared to happy faces at the short 200 ms stimulus onset asynchrony (SOA), but only when the emotional faces were presented to the left visual field. At the two longer SOAs (500 ms, 1250 ms), the high-PDI group were also faster to respond to dot-probes presented in the same location as angry compared to happy faces and slower to respond to dot-probes presented in different spatial locations to angry (vs. happy) faces. The latter effects were seen whether emotional faces were presented to the left or the right visual field. Conclusions: Results support the operation of emotion-selective engagement and defective disengagement for threat-related facial expressions (i.e., anger) in delusion-prone individuals.