We have studied geometries and rates of late Cenozoic thrust faulting and folding along the northern piedmont of the Tien Shan mountain belt, West of Urumqi, where the M=8.3 Manas earthquake occurred on December 23, 1906. The northern range of the Tien Shan, rising above 5000 m, overthrusts a flexural foredeep, filled with up to 11,000 m of sediment, of the Dzungarian basement. Our fieldwork reveals that the active thrust reaches the surface 30 km north of the range front, within a 200-km-long zone of Neogene-Quaternary anticlines. Fault scarps are clearest across inset terraces within narrow valleys incised through the anticlines by large rivers flowing down the range. In all the valleys, the scarps offset vertically the highest terrace surface by the same amount (10.2¿0.7 m). Inferring an early Holocene age (10¿2 kyr) for this terrace, which is continuous with the largest recent fans of the piedmont, yields a rate of vertical throw of 1.0¿0.3 mm/yr on the main active thrust at the surface. A quantitative morphological analysis of the degradation of terrace edges that are offset by the thrust corroborates such a rate and yields a mass diffusivity of 5.5¿2.5 m2/kyr. A rather fresh surface scarp, 0.8¿0.15 m high, that is unlikely to result from shallow earthquakes with 6<M<7 in the last 230 years, is visible at the extremities of the main fold zone. We associate this scarp with the 1906 Manas earthquake and infer that a structure comprising a deep basement ramp under the range, gently dipping flats in the foreland, and shallow ramps responsible for the formation of the active, fault propagation anticlines could have been activated by that earthquake. If so, the return period of a 1906 type event would be 850¿380 years.

The small size of the scarp for an earthquake of this magnitude suggest that a large fraction of the slip at depth (≈2/3) is taken up by incremental folding near the surface. Comparable earthquakes might activate flat detechments and ramp anticlines at a distance from the front of other rising Quaternary ranges such as the San Gabriel mountains in California or the Mont Blanc-Aar massifs in the Alps.

We estimate the finite Cenozoic shortening of the folded Dzungarian sediments to be of the order of 30 km and the Cenozoic shortening rate to have been 3¿1.5 mm/yr. Assuming comparable shortening along the Tarim piedmont and minor additional active thrusting within the mountain belt, we infer the rate of shortening across the Tien Shan to be at least 6¿3 mm/yr at the longitude of Manas (≈85.5¿E). A total shortening of 125¿30 km is estimated from crustal thickening, assuming local Airy isostatic equilibrium. Under the same assumption, serial N-S sections imply that Cenozoic shortening across the belt increases westwards to 203¿50 km at the longitude of Kashgar (≈76¿E), as reflected by the westward increase of the width of the belt. This strain gradient implies a clockwise rotation of Tarim relative to Dzungaria and Kazakhstan of 7¿2.5¿ around a pole located near the eastern extremity of the Tien Shan, west of Hami (≈96¿E, 43.5¿N), comparable to that revealed by paleomagnetism between Tarim and Dzungaria (8.6¿8.7¿).

A 6 mm/yr rate of shortening at the longitude of Manas would imply a rate of rotation of 0.45¿/m.y. and would be consistent with a shortening rate of 12 mm/yr north of Kashgar. Taking such values to be represenative of Late Cenozoic rates would place the onset of reactivation of the Tien Shan by the India-Asia collision in the early to middle Miocene (16+22/-9 m.y.), in accord with the existence of particularly thick late Neogene and Quaternary deposits. Such reactivation would thus have started much later than the collision, roughly at the time of the great mid-Miocene changes in tectonic regimes, denudation and sedimentation rates observed in southest Asia, the Himalayas and the bay of Bengal, and of the correlative rapid change in seawater Sr isotopic rate (20 to 15 Ma). Like these other changes, the rise of the Tien Shan might be a distant consequence of the end of Indochina's escape.