Signals obtained from most of real-world systems, especially from living organisms, are irregular, often chaotic, non-stationary, and noise-corrupted. Since modern measuring devices usually realize digital processing of information, recordings of the signals take the form of a discrete sequence of samples (a time series). The present paper gives a brief overview of the possibilities of such experimental data processing based on reconstruction and usage of a predictive empirical model of a time realization under study. The technique of reconstruction of mathematical models from time series is described and possibilities of the approach are illustrated with examples from the author’s and his colleagues’ experience.