AbstractA layer‐by‐layer waveform inversion method is developed to derive the velocities of a multi‐layered model. Firstly, one new concept, medium perturbation‐response, is introduced and the gradient method for waveform inversion is analysed. Essentially, waveform inversion is to extract residual structural information between the real earth and predicted model from the waveform residuals between the observed and the synthetic seismograms. The strength of waveform inversion is the high resolution of the inversion result by using a large quantity of waveform information. A weakness of waveform inversion is time‐consuming. In addition, waveform information will tend to get stuck in local minima if the starting model is too far from the actual model. The reason for this failure is that the misfit function can be highly nonlinear with respect to velocity models. The layer‐by‐layer method is to minimize the misfit function for every layer from top to bottom. In this case the total misfit function will be minimum too. For each layer, we use the dichotomy method to scan the velocity. This method can improve the speed of computation and avoid getting stuck in local minima for the iterative algorithm. Velocity and reflection interfaces can be estimated quickly and accurately based on the medium perturbationresponse and the change of the misfit function.