The interplay between modern astrometry and gravitational physics is very important for the progress in both these fields. Below some threshold of accuracy, Newtonian physics fails to describe observational data and the Einstein's relativity theory must be used to model the data adequately. Many high-accuracy astronomical techniques have already passed this threshold. Moreover, modern astronomical observations cannot be adequately modeled if relativistic effects are considered as small corrections to Newtonian models. The whole way of thinking must be made compatible with relativity: this starts with the concepts of time, space and reference systems. An overview of the standard general-relativistic framework for modeling of high-accuracy astronomical observations is given. Using this framework one can construct a standard set of building blocks for relativistic models. A suitable combination of these building blocks can be used to formulate a model for any given type of astronomical observations. As an example the problem of four dimensional solar system ephemerides is exposed in more detail. The limits of the present relativistic formulation are also briefly summarized. On the other hand, high-accuracy astronomical observations play important role for gravitational physics itself, providing the latter with crucial observational tests. Perspectives for these astronomical tests for the next 15 years are summarized.