The recent history of events can influence responding despite there being no contingent relationship between those events. These ‘sequential effects’ are ubiquitous in cognitive psychology, yet their study has been dominated by two-choice reaction time tasks in which sequences necessarily comprise simple response repetitions and alternations. The current study explored sequential effects in a three-choice reaction time task where the target was constrained to either move clockwise or anticlockwise on each trial, allowing for assessment of sequential effects involving the direction of target transitions rather than target location. Across two experiments, a reliable pattern of sequential effects was found in the absence of contingencies, whereby the most notable feature was that participants were fastest to respond to subsequences where the target moved in a consistent direction on consecutive trials, compared to when the target direction alternated. In Experiment 2, the direction of motion was biased to move in one direction 75% of the time and in a subsequent transfer phase, participants showed evidence of learning this probabilistic sequence but still exhibited the same pattern of sequential effects on trials where the target moved in the more prevalent or less prevalent direction. Simulations with a connectionist model of sequence learning (the Augmented Serial Recurrent Network, Cleeremans & McClelland, 1991) produced an adequate replication of the sequential effects in both experiments in addition to an effect of sequence learning in Experiment 2. We propose that sequential effects may represent learning about transient contingencies and may be described using the same associative learning mechanisms intended for sequence learning.