The space-time distribution of slip of the 17 August 1999 Izmit earthquake is investigated by inverting synthetic aperture radar (SAR) interferometry and Global Positioning System (GPS) data, together with teleseismic broadband and near-field strong-motion records. Surface offsets are-used as an added constraint. Special emphasis is given to analysis of the different data sets. We use a four-segment finite fault model and a nonlinear inversion scheme, allowing slip to vary in amplitude, direction, and duration, as well as variable rupture velocity. From the inversion of synthetic data, we find that the best spatial resolution can be expected in the upper half of the fault model (above 12 km), where coverage of the interferometric SAR data is good (western half of the rupture), and near the GPS and strong-motion stations. Teleseismic data are found to have a lower resolution that is more evenly distributed over the fault model. The joint inversion of all the data sets has an increased resolving power compared with the separate inversion and gives a more robust description of the space and time distribution of slip. Our study shows the importance of resolution tests in evaluating the reliability of earthquake kinematic models, and it confirms that an excellent fit of a single kind of data does not necessarily imply a good retrieval of the kinematic properties of an earthquake. The Izmit rupture, which is almost pure right-lateral strike-slip faulting, is dominated by the bilateral breaking of a central asperity located between 29.7 degrees E (about 10 km west of the city of Golcuk) and 30.4 degrees E (eastern margin of Sapanka Lake), with slip reaching 6-8 m in the depth range 6-12 km. The western termination of the rupture is found near the city of Yalova, but large slip ends around 29.7 degrees E (about 10 km east of Hersek Delta). A second area of large slip is required by all the data sets further east toward the city of Duzce, between 30.7 degrees E and 31.1 degrees E (Karadere and Duzce faults). This eastern slip zone, which is separated from the main central asperity by an area of greatly reduced slip, is less well constrained by the data. However, a strong-motion station near the city of Duzce helps to locate a high-slip patch near 31.1 degrees E in the depth range 6-12 km. The total seismic moment resulting from the joint inversion is 2.4X10 (super 27) dyne cm. Most of the energy release occurred in a short time, less than 15 sec, corresponding to the bilateral breaking of the central asperity. Rupture propagation is relatively uniform and fast toward the west, with a rupture velocity close to 3.5 km/sec. Propagation of large slip toward the east is initially slower, but it accelerates during a short time interval about 10 sec after rupture nucleation. Eastward progression then slows down to less than 2 km/sec after 15 sec, and rupture almost vanishes in amplitude ca. 20 sec after initiation. Rupture propagation then proceeds on the easternmost Karadere and Duzce fault segments, east of 30.7 degrees E, from 22 to ca. 50 sec. Supershear rupture propagation is not required for modeling the waveforms considered in this study. The hypocenter of the Duzce earthquake, which occurred 3 months later (12 November 1999, M (sub w) 7.2), is located in the immediate vicinity of the easternmost slip patch of the Izmit earthquake.