Abstract: Currently, full-waveform inversion (FWI) represents the velocity inversion and modeling technology with the highest accuracy. By continuously updating the velocity model through an iterative algorithm, FWI gradually minimizes the errors between forward modeling data and real data, thereby approaching the true velocity model. The primary challenge faced by FWI is cycle skipping, which occurs when the phase difference between forward modeling data and real data exceeds half a cycle. In this case, with the updating of the velocity model, the forward modeling data and real data will be incorrectly matched, causing the velocity model to converge in a wrong direction. In conventional least-squares FWI, the objective function considers merely the energy difference between forward modeling data and real data. In this case, cycle skipping can only be avoided as far as possible by using the low-frequency signals with high cycles in the data and a sufficiently accurate initial velocity model. This poses stringent demands for the quality of low-frequency signals in raw data and entails a substantial workload for the initial velocity modeling. To address this issue, this study proposes a time-lag FWI (TLFWI) method. In this method, the objective function is defined as the integral of the energy difference between forward modeling data and real data with respect to their time difference. In this manner, the objective function accounts for both the energy difference and time difference between forward modeling data and real data, enabling cycle skipping to be effectively avoided. In the case of weak low-frequency signals in raw data and large errors in the initial velocity model, TLFWI can still enable the velocity model to converge in the correct direction. Therefore, TLFWI reduces the dependence on low-frequency signals in raw data and decreases the workload for constructing the initial velocity model. The results of experiments on a theoretical velocity model and practical seismic data reveal that, compared to least-squares FWI, TLFWI can derive a more accurate velocity model under the condition of high errors in the initial velocity model. Notably, both least squares FWI and TLFWI used in this study are turning wave FWI. In other words, only turning waves in seismic data are used for matching in both methods, with reflected waves excluded.