In this experiment we investigated the effect of different instructions on the modification of attentional biases, and subsequently on worry persistence. Participants without excessive worry completed a modified dot-probe task, designed to train attention either to threat or neutral words. Half of each group was given explicit instructions regarding the relationship between word valence and target location, and half were given the more usual minimal instructions. Impact on worry persistence was assessed by categorizing the valence of thought intrusions before and after a period of instructed worry. Response latencies to test items on the dot-probe task showed that attention had been successfully manipulated in the expected direction, and explicit instructions led to more effective attention modification. Moreover, participants in the attend-threat group who received explicit instructions reported significantly more negative thought intrusions following instructed worry, as rated by an assessor, whereas participants in the attend-neutral group did not. These findings suggest that an attentional bias towards threatening information plays a role in worry persistence, and that explicit instructions may be helpful in modifying this bias.