Little is known about how observers' scanning strategies affect performance when monitoring events in closed-circuit television (CCTV) footage. We examined the fixation behaviour of change detectors and non-detectors monitoring dynamic scenes. One hundred forty-seven participants observed mock CCTV videos featuring either a mock crime or no crime. Participants were instructed to look for a crime, to look for something unusual or simply to watch the video. In both videos, two of the people depicted switched locations. Eye movements (the number of fixations on the targets and the average length of each fixation on targets) were recorded prior to and during the critical change period. Change detection (24% overall) was unaffected by event type or task instruction. Fixation behaviour differed significantly between the criminal and non-criminal event conditions. There was no effect of instructions on fixation behaviour. Change detectors fixated for longer on the target directly before the change than did non-detectors. Although fixation behaviour before change predicted change detection, fixation count and durations during the critical change period did not. These results highlight the potential value of studying fixation behaviour for understanding change blindness during complex, cognitively demanding tasks (e.g. CCTV surveillance).