Palaeomagnetic studies of the Chinese loess provided the first firm chronology for these sediments and revealed that their depositional history reaches back over 2.5 Ma. Magnetic susceptibility records can provide an even more detailed chronology through correlation with the δ18O deep-sea stratigraphy. The susceptibility fluctuations have been used also, either alone or in combination with 10Be measurements, to reconstruct palaeoclimate. However, there is still debate as to the origin of the 10Be and magnetic minerals in the Chinese loess and hence their palaeoclimatic significance. Here we analyse magnetic data from six sites across the Loess Plateau, and reinterpret the 10Be data from Luochuan as largely reflecting primary 10Be dust loading rather than as a palaeorainfall indicator. We discount previous ideas of both constancy of the flux of magnetically susceptible minerals (i.e. susceptibility flux) or constancy of the ratio of the flux of 10Be and susceptibility. We explain the susceptibility of Chinese loess in terms of in situ pedogenic ferrimagnetic enhancement. The competing effects of reduction and oxidation are taken to control the degree of magnetic enhancement, such that magnetic susceptibility is a direct indicator of palaeorainfall. Our reconstructions yield variations in palaeorainfall of between +28 per cent and −12 per cent in the central Loess Plateau rather than +100 per cent to −50 per cent as suggested from previous 10Be and magnetic susceptibility studies.