Objectives. The objective of this study was to investigate how the characteristics of self-generated and spontaneous images might differ between states of high and low anxiety, as observed in spider phobia. It was not known whether self-generated images could be used, in the same way as spontaneous images, to access core beliefs. The frequency and characteristics of spontaneous images were assessed to determine whether they are negative, recurrent, and link to early memories, as have been reported in social phobia. Design. Participants were recruited into either a spider-anxious group or a control group based upon their response (independent variable) to the Fear of Spiders Questionnaire (Muris & Mercklebach, 1996). A semi-structured interview, incorporating a within and between group design, required participants to rate the characteristics (the dependent variables) of self generated and spontaneous images. Method. A semi-structured interview was administered. Participants rated the characteristics of two images (spider and butterfly) using visual analogue scales. The downward arrow technique was used to access core beliefs associated with the self-generated and spontaneous images. Results. The spider-anxious group's spider image was more vivid, evoked more anxiety, and was perceived as having more intent than both control images. The spider-anxious group reported more negative core beliefs associated with the self-generated image and more spontaneous images, that were recurrent, negative and that linked to early traumatic experiences. Conclusions. The phenomenological characteristics of self-generated images can be reliably assessed and, in the absence of spontaneous images, can be utilised to access core beliefs in anxiety disorders. This study provides some evidence of cognitive biases and thinking errors and calls for a new cognitive model of specific phobia.