The correct interpretation of full-space transient electromagnetic data has always constituted a critical safety problem during tunnel excavation projects. Targeting the interpretation of water-filled caves under narrow tunnel conditions, we have developed an iterative inversion method based on 3D finite-difference time-domain (FDTD) forward calculations and a direction algorithm. In total, 125 groups of 3D FDTD forward calculation results are analyzed to identify the correlations between the response data and the geometric conditions of the cave. A direction algorithm is established based on the correlations, thereby increasing the iterative inversion convergence speed. Using the proposed iterative inversion method, the location and volume of the water-filled cave in front of the tunnel face are successfully inverted. Through an iterative program, the inversion results of simulations involving the detection of water-filled caves under tunnel conditions are accurately analyzed, and the relative error is less than 10%. The application of the iterative inversion method to the Mingyue Mountain Tunnel project suggests that this method is capable of interpreting the size of water-filled caves and it is valid for a narrow tunnel face with only a single available measurement point. The proposed iterative inversion method can be used alone or in combination with other detection techniques, thereby providing engineers with a better early warning system for detecting water-filled caves in tunnels.