We present a survey to a new method of analyzing experimentally gained stochastic time series. The method is based on the theory of Markov processes and does not depend on previous knowledge of model equations. An overview about the underlying technique and several applications are given. We show for the complicated case of a chaotic dynamics spoiled at the same time by dynamical and measurement noise how to extract from data the magnitude of both types of noise. © 2004 American Institute of Physics