Abstract: The convergence speed of Genetic Algorithms is related to the devision precision of original search scope of model parameters. The improvement has been put forward in this paper, which allows the iteration to begin at low precision, and reduces the search scope progressively in iteration. So that it makes the convergence speed of Genetic algorithms and inversion precision raise simultaneouslyin iteration, and does not extend the dimension of model parameters space. The inversed results of I-D crustal elocity structure has proved that the method is efficient. The tactics used in practice is also presented.