Background and objectives: This research compared the effects of real versus hyper-real images on anxiety, disgust, and physiological arousal during internet-delivered exposure in high spider-fearfuls. Hyper-real images were digitally altered to highlight fearful aspects. A further aim was to examine self-reported and behavioural therapeutic outcomes and exposure-related predictors of these outcomes. Methods: Twenty-eight females were randomised to real (n = 14) or hyper-real (n = 14) treatment groups and nine participants were subsequently allocated to a wait-list control group. Treatment groups viewed an 8-stage exposure hierarchy of real or hyper-real spider images. Subjective anxiety and disgust ratings were taken during each stage (0, 60, 120, 180 s) with heart rate and skin conductance recorded throughout. Results: Anxiety, disgust and physiological arousal habituated within each exposure stage, with no differential effect of real compared to hyper-real images. Both treatment groups but not controls demonstrated significant reductions in behavioural avoidance and self-reported phobic symptoms from pretreatment to post-treatment with large effect sizes noted. The change in within-stage habituation of anxiety, disgust and heart rate, between the first and last stage, predicted improvement in behavioural avoidance at post-treatment. This suggests that generalisation of habituation to multiple images is an important predictor of improvement. Limitations: While findings in relation to therapeutic outcome should be considered preliminary, clear relationships were found between exposure-related variables and outcome among those who undertook treatment. Conclusions: Findings provide evidence in support of the efficacy of online image-based exposure and have implications for informing further research into the underlying mechanisms of image-based exposure treatment.