This paper reports an experimental investigation of engagement with and disengagement from a threatening cue of pain. As most paradigms in pain research only provide an overall index of attentional deployment by pain-related information, a new paradigm was developed that allowed an independent investigation of engagement with and disengagement from pain cues. Forty pain-free volunteers performed a cueing task in which they had to detect pain targets and tone targets as quickly and as accurately as possible. The target stimuli were preceded by pain cues (the word 'pain'), tone cues (the word 'tone'), or neutral cues (a series of the character 'X') at stimulus onset asynchrony (SOA: stimulus interval between cue onset and target onset) levels of 100, 500, or 900 ms. There was no contingency between the type of cue and the type of target. Catastrophic thinking about pain and the predictive value of the cues were assessed by self-reports. Results can be summarized as follows: When a cue correctly primed a target, attention was optimally engaged in the identification of the target irrespective of the threatening context of the cue or target. However, when pain was cued and did not occur, there was retardation in disengagement from the pain cue. This retardation was more pronounced and extended across time in those high in catastrophic thinking about pain. On examination it appeared that catastrophic thinking about pain may operate by a protection of the belief that the cue for pain is a valid one, despite experience to the contrary.