Using a serial reaction time task (Nissen & Bullemer, 1987), two experiments directly examined people's ability to implicitly learn first- (Experiment 1) and second-order (Experiment 2) transition probabilities. On each trial, the target appeared at one of four locations on the screen and participants pressed the corresponding key. In experiment 1, the probability that the target appeared at a particular location on trial (t) given its location on trial (t-1) was 0, 0.40, 0.50, or 0.60. In experiment 2, the probability that the target appeared at a particular location on trial (t) given its locations on trials (t-2) and (t-1) was 0, 0.40, 0.50, or 0.60. In both experiments, reaction time (error rate) was slower (greater) on low (0.40) than high (0.60) probability transitions. Perfor ance on medium (0.50) transitions was in between. Differences in reaction time and error rate were greater in experiment 1 than experiment 2. The results suggest first- and second-order probabilities were learned, but that learning of second-order probabilities was impaired relative to learning of first-order probabilities. On explicit measures asking participants to indicate which transitions were more likely to occur (Experiment 1) or to estimate the transition probabilities (Experiment 2), performance was at chance. The results provide strong evidence that people can implicitly learn first- and second-order transition probabilities.