Caesium magnetometers are new tools for soil property mapping with a decimetric resolution [Mathé, V., Lévêque, F., 2003. High resolution magnetic survey for soil monitoring: detection of drainage and soil tillage effects. Earth and Planetary Science Letters 212 (1–2), 241–251]. However, when the magnetic anomalies are only a few nanoteslas (nT), the geologic and pedogenic signal must first be isolated from magnetic disturbances for this method to be useful. This paper investigates the instrumental artifacts and environmental disturbances to adapt the survey protocol to slightly magnetic soils. Among the possible instrumental sources of disturbances listed and quantified, the most significant are: 1) The battery effect upon sensors 2 m away (classic protocol, about ± 0.15 nT) while increasing this distance up to 10 m cancelled it; 2) The noise level of magnetometers and sensors, which, according to tests on two magnetometers and three sensors, rarely and randomly exceeds 0.1 nT, but seems to increase with the electronic component age. Among the environmental disturbances, temporal variations such as diurnal variation or fluctuations linked to the moving of metallic masses play a major role, although the pseudogradient or base-station methods have commonly cancelled them. The efficiency of the latter is strongly dependent on the source nature. However, the ground currents and electromagnetic fields propagating in soils cause more problems. As a first step to better understand such disturbance sources, uncommon magnetic signal variations supposedly due to electromagnetic wave conversions and likely linked to the railway traffic are presented. Based on previous results, an adapted protocol using one magnetometer and two caesium sensors (0.3 and 1.6 m above the surface) is proposed to increase the signal / noise ratio. At first, to maintain an accurate horizontal and vertical location of the sensors, the latter are affixed to a wooden handcart running on plastic rails. Rails adapt to micro-topography, thereby decreasing strongly the soil–sensors distance variations. Anomalies due to topography rarely exceed 0.1 nT. Finally, a method to remove diurnal variations from high-resolution magnetic maps is proposed. Parallel profiles performed successively are adjusted by a cross-profile. Assuming that the temporal variations during each profile are negligible (less than 0.05 nT), this technique, contrary to the pseudogradient, preserves both the decimetric and the metric anomalies (gain of more than 1 nT)