Generalised avoidance behaviours are a common diagnostic feature of anxiety-related disorders and a barrier to affecting changes in anxiety during therapy. However, strategies to mitigate generalised avoidance are under-investigated. Even less attention is given to reducing the category-based generalisation of avoidance. We therefore investigated the potential of an operant-based approach. Specifically, it was examined whether reinforcing competing (non-avoidance) behaviours to threat-predictive cues would interfere with the expression of generalised avoidance. Using a matching-to-sample task, artificial stimulus categories were established using physically dissimilar nonsense shapes. A member of one category (conditioned stimulus; CS1) was then associated with an aversive outcome in an Acquisition context, unless an avoidance response was made. Next, competing behaviours were reinforced in response to the CS1 in new contexts. Finally, we tested for the generalisation of avoidance to another member of the stimulus category (generalisation stimulus; GS1) in both a Novel context and the Acquisition context. The selective generalisation of avoidance to GS1 was observed, but only in the Acquisition context. In the Novel context, the generalisation of avoidance to GSs was significantly reduced. A comparison group (Experiment 2), which did not learn any competing behaviours, avoided GS1 in both contexts. These findings suggest that reinforcing competing behavioural responses to threat-predictive cues can lead to reductions in generalised avoidance. This study is among the first study to demonstrate sustained reductions in generalised avoidance resulting from operant-based protocols, and the clinical and research implications are discussed.