We present empirical evidence that human reasoning follows the rules of probability theory, if information is presented in “natural formats”. Human reasoning has often been evaluated in terms of humans’ ability to deal with probabilities. Yet, in nature we do not observe probabilities, we rather count samples and their subsets. Our concept of Markov frequencies generalizes Gigerenzer and Hoffrage’s “natural frequencies”, which are known to foster insight in Bayesian situations with one cue. Markov frequencies allow to visualize Bayesian inference problems even with an arbitrary number of cues.