The brain needs to activate multiple networks in a temporally coordinated manner in order to perform cognitive tasks, and it needs to do so at different temporal scales, from the slowest cyrcadian cycles to fast subsecond rhythms. When trying to characterise these processes, our choice of methods has an impact on the way we understand brain function. I will explain why is that, and, bearing this in mind, I will introduce a probabilistic framework to investigate brain functional reorganisation, capable to reliably access the dynamics contained in the signal even at the fastest time scales. I will review some studies on the intrinsic dynamics of the human brain that we investigated using this approach. I will then show how these models are associated, or can be associated, to behaviour at different levels.