Abstract: As an observation science, high-density resistivity method is likely to be disturbed by external noise signal during field data acquisition, causing distortion of the current line and false distortion of the non-geological structure. Aiming at the recorded resistivity data which was the compound field superposed signal mixed with noise, we introduce three methods of data space correction, model space correction and robust inversion to correct the geo-electric data with noise. The results show that the three methods all can inhibit the influence of random disturbance noise to a certain extent, remove the false signal and obtain the true data. Compared with the former two methods, the image anomaly pattern obtained from the robust inversion is more regular and convergent. Furthermore, the boundary between the low resistance zone and the surrounding medium is clear, and the result is more reliable. Based on the analysis of the actual leakage detection data of a dam, it is proved that the robust inversion has a better effect of inhibiting noise without the need of establishing the initial geo-electric model and can provide a new solution for the processing of low signal-noise ratio data.