In the antisaccade task participants are required to overcome the strong tendency to saccade towards a sudden onset target, and instead make a saccade to the mirror image location. The task thus provides a powerful tool with which to study the cognitive processes underlying goal directed behaviour, and has become a widely used index of "disinhibition" in a range of clinical populations. Across two experiments we explored the role of top-down strategic influences on antisaccade performance by varying the instructions that participants received. Instructions to delay making a response resulted in a significant increase in correct antisaccade latencies and reduction in erroneous prosaccades towards the target. Instructions to make antisaccades as quickly as possible resulted in faster correct responses, whereas instructions to be as spatially accurate as possible increased correct antisaccade latencies. Neither of these manipulations resulted in a significant change in error rate. In a second experiment, participants made fewer errors in delayed pro and antisaccade tasks than in a standard antisaccade task. The implications of these results for current models of antisaccade performance, and the interpretation of antisaccade deficits in clinical populations are discussed.