Most of Kazakhstan belongs to the central part of the Eurasian Paleozoic mobile belts for which previously proposed tectonic scenarios have been rather disparate. Of particular interest is the origin of strongly curved Middle and Late Paleozoic volcanic belts of island-arc and Andean-arc affinities that dominate the structure of Kazakhstan. We undertook a paleomagnetic study of Carboniferous to Upper Permian volcanics and sediments from several localities in the Ili River basin between the Tien Shan and the Junggar–Alatau ranges in southeast Kazakhstan. Our main goal was to investigate the Permian kinematic evolution of these belts, particularly in terms of rotations about vertical axes, in the hope of deciphering the dynamics that played a role during the latest Paleozoic deformation in this area. This deformation, in turn, can then be related to the amalgamation of this area with Baltica, Siberia, and Tarim in the expanding Eurasian supercontinent. Thermal demagnetization revealed that most Permian rocks retained a pretilting and likely primary component, which is of reversed polarity at three localities and normal at the fourth. In contrast, most Carboniferous rocks are dominated by postfolding reversed overprints of probably “mid-Permian” age, whereas presumably primary components are isolated from a few sites at two localities. Mean inclinations of primary components generally agree with coeval reference values extrapolated from Baltica, whereas declinations from primary as well as secondary components are deflected counterclockwise (ccw) by up to ∼ 90°. Such ccw rotated directions have previously also been observed in other Tien Shan sampling areas and in the adjacent Tarim Block to the south. However, two other areas in Kazakhstan show clockwise (cw) rotations of Permian magnetization directions. One area is located in the Kendyktas block about 300 km to the west of the Ili River valley, and the other is found in the Chingiz Range, to the north of Lake Balkhash and about 400 km to the north of the Ili River valley. The timing of the ccw as well as cw rotations is clearly later than the disappearance of any marine basins from northern Tarim, the Tien Shan and eastern Kazakhstan, so that the rotations cannot be attributed to island-arc or Andean-margin plate settings — instead we attribute the rotations to large-scale, east–west (present-day coordinates), sinistral wrenching in an intracontinental setting, related to convergence between Siberia and Baltica, as recently proposed by Natal'in and Şengör [Natal'in, B.A., and Şengör, A.M.C., 2005. Late Palaeozoic to Triassic evolution of the Turan and Scythian platforms: the pre-history of the palaeo-Tethyan closure, Tectonophysics, 404, 175–202.]. Our previous work in the Chingiz and North Tien Shan areas on Ordovician and Silurian rocks suggested relative rotations of ∼ 180°, whereas the Permian declination differences are of the order of 90° between the two areas. Thus, we assume that about 50% of the total post-Ordovician rotations are of pre-Late Permian age, with the other half of Late Permian–earliest Mesozoic age. The pre-Late Permian rotations are likely related to oroclinal bending during plate boundary evolution in a supra-subduction setting, given the calc-alkaline character of nearly all of the pre-Late Permian volcanics in the strongly curved belts.