Construction of mathematical models from experimental data is a topical field in mathematical statistics and nonlinear dynamics. It has a long history and still attracts increasing attention. We briefly discuss key problems in nonlinear modeling for typical problem settings ("white", "gray", and "black boxes") and illustrate several contemporary approaches to their solution with simple examples. Finally, we describe a technique for determination of weak directional coupling between oscillatory systems from short time series based on empirical modeling of their phase dynamics and present its applications to climatic and neurophysiological data.