To understand how people build probabilistic internal representations of their dynamic perceptual environment, it is essential to know how the statistical structures of event sequences are encoded in the brain. Previous attempts either characterized this coding by the structure of short-term repetition/alternations or while acknowledging the importance of long-term baseline probabilities, they failed to explore their effect by manipulating properly the baseline statistics. We investigated how expectations about the probability of a visual event are affected by varying short-term and unbalanced baseline statistics. Participants (N=19) observed sequences of visual presence-absence events and reported about their beliefs by two means: by quickly pressing a key indicating whether or not an object appeared and by giving interspersed numerical estimates of the appearance probability of the event together with their confidence of their answer. Stimuli appeared at random with the baseline probabilities systematically manipulated throughout the experiment. We found that reaction times (RTs) for visual events did not depend exclusively on short-term patterns but were reliably influenced by the baseline appearance probabilities independent of the local history. Error rates, RTs and explicit estimates were similarly influenced by the baseline: subjects were more accurate estimating the probability of very likely and very unlikely events. Furthermore, we found that subjects’ report of their confidence was systematically related to both the implicit and explicit accuracy measures. Finally, reaction times could be explained by a combined effect of short-term and baseline statistics of the observed events. These results indicate that the perception of probabilistic visual events in a dynamic visual environment is influenced by short-term patterns as well as automatically extracted statistics acquired on the long run. Our findings lend support to proposals that explain behavioral changes in terms of relying on an internal probabilistic model rather than as a local adaptation mechanism.