Self-potential signals and resistivity data can be jointly inverted or analyzed to track the position of the burning front of an underground coal-seam fire. We first investigate the magnitude of the thermoelectric coupling associated with the presence of a thermal anomaly (thermoelectric current associated with a thermal gradient). A sandbox experiment is developed and modeled to show that in presence of a heat source, a negative self-potential anomaly is expected at the ground surface. The expected sensitivity coefficient is typically on the order of in a silica sand saturated by demineralized water. Geophysical field measurements gathered at Marshall (near Boulder, CO) show clearly the position of the burning front in the electrical resistivity tomogram and in the self-potential data gathered at the ground surface with a negative self-potential anomaly of about . To localize more accurately the position of the burning front, we developed a strategy based on two steps: (1) We first jointly invert resistivity and self-potential data using a cross-gradient approach, and (2) a joint interpretation of the resistivity and self-potential data is made using a normalized burning front index (NBI). The value of the NBI ranges from 0 to 1 with 1 indicating a high probability to find the burning front (strictly speaking, the NBI is, however, not a probably density). We validate first this strategy using synthetic data and then we apply it to the field data. A clear source is localized at the expected position of the burning front of the coal-seam fire. The NBI determined from the joint inversion is only slightly better than the value determined from independent inversion of the two geophysical data sets.