We present the results of a paleomagnetic study of 360 cores, drilled at three different areas from the Tibetan Plateau: 13 sites from Lower Cretaceous red beds in the Xining-Lanzhou basin, around Xining and Lanzhou cities, southwest of the Qilian mountains (36.2 ¿N, 103.5 ¿E); 13 sites from Cretaceous red beds in the Kunlun block, near Maqin (34.5 ¿N, 100.1¿); and 9 sites from Paleocene/lower Eocene red beds in the Qiangtang block, near Fenghuoshan (34.5 ¿N, 92.8 ¿E). Thermal demagnetization of the samples allowed us to isolate a high-temperature component which passes both positive reversal and fold tests for all formations. The corresponding paleopoles lie at 50.3 ¿N, 195.5 ¿E (A95=4.6¿) for Xining-Lanzhou, 80.1 ¿N, 281.2 ¿E (dp=7.8¿, dm=12.7¿) for Maqin, and 62.6 ¿N, 210.5 ¿E (dp=3.9¿, dm=6.8¿) for Fenghuoshan. We discuss these in the frame of a new paleogeographic reconstruction of the Cretaceous paleoposition of the blocks forming the Asian mosaic. We conclude that the Xining-Lanzhou area could not be part of the North China Block but rather was associated with the Tarim-Qaidam assemblage. Paleomagnetic data argue in favor of a Qaidam-Kunlun-Tarim-Junggar assemblage in the Cretaceous, significantly to the south of its current position with respect to the Asian continent (Siberia, Mongolia and North China). The large N-S convergence (800¿500 km) implied since the Cretaceous appears to be far larger than could be absorbed in the Altay range to the north and Qilian Shan to the east (of the order of 300 km). Part of this motion could have occurred along a large left-lateral strike slip fault system, which may connect with the Mongol-Okhotsk suture to the northeast. Verifying this hypothesis will require new geologic and paleomagnetic data from these remote regions. ¿ 1998 American Geophysical Union