[1] Inconsistent stratigraphic positions of reversal boundaries and spurious paleomagnetic variations associated with polarity reversals recorded by loess/paleosols on the Chinese Loess Plateau (CLP) have in general been attributed to inferred delayed acquisition of magnetization and high-frequency variations of the geomagnetic field, respectively. These conclusions are partly based on an incomplete knowledge of the magnetization acquisition process of loess. Recent laboratory experiments, which demonstrate that dry-deposited loess acquires a stable remanent magnetization after its first wetting (precipitation), resulted in significant inclination errors that are not generally encountered in Chinese loess. Laboratory experiments only remotely mimic natural conditions, and we have therefore performed in situ experiments at two sites on the CLP separated by some 650 km to elucidate whether variations in natural conditions can impose secondary remanent magnetization components in loess (L1). The experiments were performed by relocating blocks (volume ∼5·10−3 m3) of last interglacial loess after being turned upside down inside loess exposures. Inverted blocks were retrieved and analyzed after 8 and 17 months respectively. Thermal demagnetization reveals that partial remagnetization has occurred within different parts of the loess blocks. This secondary magnetization, which is characterized by blocking temperatures that overlap with those of the primary component to above 450°C, cannot be attributed to a viscous remanent magnetization and is therefore proposed to represent local and partial realignment of magnetic grains. We tentatively propose that physical ‘unlocking’ and realignment of magnetic grains may be assisted by pressure release caused by differential thermal expansion of the different mineral grains in loess. Localized remagnetization may account for a number of reported spurious paleomagnetic results from CLP.