The problem of determination of the orbital velocity of an astrometric satellite from its own observational data is studied. It is well known that data processing of microarcsecond-level astrometric observations imposes very stringent requirements on the accuracy of the orbital velocity of satellite (a velocity correction of 1.45 mm/s implies an aberrational correction of 1 μas). Because of a number of degeneracies the orbital velocity cannot be fully restored from observations provided by the satellite. Seven constraints that must be applied on a velocity parameterization are discussed and formulated mathematically. It is shown what part of velocity can be recovered from astrometric data using a combined fit of both velocity parameters and astrometric parameters of the sources. Numerical simulations show that, with the seven constraints applied, the velocity and astrometric parameters can be reliably estimated from observational data. It is also argued that the idea to improve the velocity of an astrometric satellite from its own observational data is only useful if a priori information on orbital velocity justifies the applicability of the velocity constraints. The proposed model takes into account only translational motion of the satellite and ignores any satellite-specific parameters. Therefore, the results of this study are equally applicable to both scanning missions similar to Gaia, and pointing ones like SIM, provided that enough sources were observed sufficiently uniformly.