Climate oscillations may strongly modify continental precipitation and temperature patterns, therefore understanding their history is relevant for comprehending effects of past and ongoing climate changes. For this purpose, temperature and precipitation reconstructions beyond the instrumental record are extremely useful. As widespread terrestrial archives, loess–paleosol sequences are viable targets for such analyses. Consequently, cost-efficient geophysical proxies have gained increasing attention, but little is known about their capability to reflect even narrow climatic differences. Here we assess the sensitivity of rock-magnetic and photo-spectrometric properties of topsoil samples (n = 50) along uncorrelated, mean annual precipitation (MAP: 525±1 mm/yr to 584±1 mm/yr) and mean annual temperature (MAT: 10.8±0.1 °C to 11.2±0.1 °C) gradients across the Bačka Loess Plateau (Serbia) and test a multivariate approach. Most proxies are sensitive to MAP <565±1 mm/yr, especially anhysteretic remanent magnetization (r2 = 0.81). Applying a multivariate approach to hysteresis data reveals a robust relationship between precipitation (r2 = 0.63), aridity (r2 = 0.67) and physical properties over the entire MAP range. Although the approach needs to be further tested considering different climates, regression analyses, and timescales, our study indicates that multi-proxy approaches may increase the robustness with respect to single-proxy measurements for MAP and aridity reconstructions.