Abstract: Traditional geostatistical inversion usually uses geostatistics simulation algorithms to construct the prior information of the model parameters，and then uses some optimization algorithms to obtain the posterior solutions of the model parameters under the constraints of seismic data， which ignores the influence of lithology on model parameters and requires a lot of calculation in the optimization process. Therefore，the prior distribution of model parameters was expressed as Gaussian mixture distribution influenced by discrete lithology，then the linear Gaussian mixture inversion theory was combined with sequential simulation of geostatistics，and the posterior solutions of the model parameters and discrete lithology were directly obtained by sequential sampling，which avoids the optimization process，and the inversion results have good spatial continuity and stability. Both model tests and the application of actual data show the effectiveness of the method.

Key words: geostatistics, sequential simulation, Gaussian mixture, discrete lithology