Abstract: In this paper, the general theory of the source parameter inversion from geodetic data is considered to be consisted of modeling of the earthquake source, the realization of optimization and the utilization of the geodetic data. Mainly the latter two are studied in detail. A rigorous method of direct inversion from the original observation data (e.g. variations in length, angle, elevation difference, tilt and strain on the earth's surface) is put forward. The modified simple method accelerates the convergence of iteration. The convergence tests are given. The normalization of observation data according to sequential approximation guarantees that data of different observations are of the same normal distribution.The source parameters of the 1976 Tangshan earthquake are inversed from geodetic data. The stability and reliability of the results are tested. The results show right lateral slip of 2.51m and vertical slip of 0.70m (southeast side down) on a fault plane which dips 90. The strike, width and upper margin of the earthquake fault are N56E, 112 km, 15 km and 0 km respectively. The middle point position of the fault is N3936'.6, E11811',4. The seismic moment, stress drop and strain drop are 1.45 1027 dyne, cm, 37.2 bar and 5.6 10-5 respectively. This result is nearly the same as that deduced from seismic waves and other geophysical data observed during the earthquake. Therefore, it is concluded that there was rare possibility that fault creep, comparable with the faulting of the main shock, could have taken place both before and after the earthquake.