Abstract: To achieve refined inversion outcomes of TEM and enhance the precision of geological problem interpretation, conducting research on inversions under different initial models holds substantial significance. This study selected two distinct initial models: the homogeneous half-space model and the model derived from the inversion result of the least-squares method with a smoothing factor. Initially, the forward-response data of a three-layer geoelectric model (H-type and K-type) were utilized to conduct inversion computations via the least-squares method with a smoothing factor and the regularization-damped least squares method under different initial models. The inversion performances under various initial models were then compared and analyzed. Subsequently, field - measured data were employed to perform inversion calculations identical to those of the forward - fitting data. The normalized potential of the secondary field at a single point with different decay times, resistivity, the result model, and their corresponding errors were selected for comparative analysis. Additionally, the inversion effect of a complete apparent-resistivity profile was analyzed in conjunction with geological data. Finally, the research findings were validated through engineering cases. The results show that: The inversion of the least - squares method with a smoothing factor is poorly fitted with transient electromagnetic data. Under equivalent conditions, although the regularization-damped least squares method can more realistically reflect the stratal characteristics, its depiction of deep - stratum information is less accurate compared to when the initial model is the inversion - result model of the least - squares method with a smoothing factor. The inversion result of the regularization-damped least-squares method, with the inversion result of the least-squares method with a smoothing factor serving as the initial model, provides a more precise reflection of the underground geological conditions. The choice of different initial models exerts a remarkable influence on the inversion effect of TEM data. Through comparative analysis, it is further verified that the inversion of the regularization-damped least-squares method, with the inversion result of the least-squares method with a smoothing factor as the initial model, enjoys significant advantages in terms of accuracy and reliability. This conclusion offers crucial references and practical guidance for the application of the TEM in detecting complex geoelectric structure.