Studies on implicit sequence learning have employed the methodology oftask dissociations to show that tasks of conscious memory fail to reveal knowledge expressed in performance measures of learning. One critical requirement of this methodology is that the conscious memory tests tap the same type of information that is expressed in performance measures. When a deterministic sequence is being repeated during practice, identification of the exact type of sequential information that is learned can be achieved by a trial-by-trial comparison between the practised sequence and a control sequence. In Experiment 1 we examine whether short-term sequential effects are present in choice response time tasks and may therefore contaminate this trial-by-trial comparison. In Experiment 2 we control for these effects and demonstrate how specification of the exact sub-parts of the sequence that are learned is necessary before testing for task dissociations. Our findings indicate a dissociation between a response time task and a free generation task. This dissociation, however, is obtained for selected sub-parts of the sequence and may be caused by the insensitivity of the free-generation task to low confidence knowledge.