Neural representations of statistical probability and uncertainty are critical for brain function and behavior (Ma and Jazayeri 2014), and it is particularly important that the brain learn temporal relationships since they can infer causation and allow predictions about future events. Influential models assume this information is available in the brain, but many details are speculative and direct evidence remains elusive. Multiple labs (Bear, Berry, Buonomano, Buschman, Carandini, Fiser, Gavornik, Hawkins, Keller, Mrsic-Flogel, Hussain Shuler, Shouval) have demonstrated that primary visual cortex (V1) circuits are capable of reporting and predicting temporal relationships, though the degree to which statistical predictability in the environment shapes these responses is unclear. To address this issue we recorded visually-evoked activity in mouse V1 during exposure to sequences of visual stimuli presented with specific transition probabilities. Using local field potentials in layer 4 and two-photon calcium imaging in layer 2/3, we found that unexpected transitions drive prediction-error (PE) responses coding for stimulus identity and event likelihood conditioned on the recent past. We describe the speed at which PEs emerge within V1 and show that PEs, reflected in firing rates and population size, scale with the degree of uncertainty and over a wide range of probabilities. The findings demonstrate that cortical circuits rapidly encode information similar to a Bayesian prior, and may be useful for guiding the development and refinement of predictive coding theories by describing how neural activity adapts to reflect specific elements of observed statistics in the environment.