The occurrence of processing bias manifested by the modified Stroop task does not require that the subjects to be aware of the stimuli presented. Earlier studies have shown that even when stimuli are backwardly masked so conscious identification is prevented, patients suffering from Generalised Anxiety Disorders slow down colour-naming masks that are preceded by threatening words. In non-patient samples, processing bias on the modified Stroop task is related to the level of trait anxiety. We tested whether this preconscious processing bias is related to anxiety per se and whether it also occurs in specific phobias. Indeed, in a group of 37 spider phobics, the intensity of phobic complaints was significantly associated with interference measures on both the masked and the unmasked modified Stroop task. Preconscious processing bias was not associated with treatment gain. Interference on the masked and unmasked Stroop task was reduced after treatment. Though the lack of a no treatment control group preludes definite conclusions, our findings suggest that preconscious biases are influenced by behaviour therapy. Results are critically discussed.