More than 70?km of old masonry tunnels covering natural streams were recently inventoried in the Cevennes mountain region (France, département du Gard). Due to their advanced state of degradation and their shallow depth, the covered stream tunnels have been responsible for several surface collapses since 2012. The mapping of the trajectory of these tunnels therefore is of primordial importance for the risk assessment of the affected municipalities. But only few of them were mapped so far due to their difficult accessibility and their precarious stability. Consequently there is a need to identify non-destructive methods for detecting them quickly from the surface of the most preoccupying areas. Two ERT profiles using 48 electrodes and the Wenner-Schlumberger array, under dry and wet conditions, were realized at the surface of one tunnel of which the position was precisely known in order to determine the ability of the technique to detect it. Analysis of the measured apparent resistivity showed an unexpected local decrease of resistivity near the tunnel boundary. The inverse models performed without any filtering of the measured apparent resistivity did not allow detecting any anomaly that could be interpreted as the presence of the tunnel. Conversely, the inversion of a smoothed version of the apparent resistivity allows detecting a low resistivity anomaly at the tunnel position. This result was quite surprising because we rather expected a high resistivity anomaly due to the infinite resistivity of the tunnel atmosphere compared to the surroundings. According to a study recently published by Putiška et al. (2012) on this topic and to a set of forward resistivity models we conducted, this unexpected low resistivity anomaly was explained to be due to the presence of a thin conductive layer around the tunnel. The fact that this anomaly was unobserved after inversion of the unfiltered apparent resistivity was explained by the presence of a high resistivity material (probably composed of coal) beside the thin conductive layer that altered the least-squares inversion process.