Within a statistical context, we present an algorithmic approach for identifying the essential dynamical behaviour of Markovian systems like, e.g., the (high-friction) Langevin equation or the Hamiltonian dynamics with randomized momenta. The basic idea is to exploit spectral information of a certain transfer operator that is associated to the Markovian system and that represents the evolution on a macroscopic level. We describe the mathematical framework, the algorithmic idea and exemplify the approach in application to a small molecule as well as different model systems encountered in molecular dynamics.