Abstract. Variation function is an important tool that can characterize the spatial correlation of geological variable quantitatively in geo-statistics. In the random inversion process, for different types of blocks and oil fields, choosing the appropriate function type and the corresponding model parameters, to improve the accuracy of inversion results is particularly important. By taking Gu83 block of GuLong oil field as an example, this article analyses common variation functions such as Gaussian, Spherical, Expo and Gen-Expo method and the influence of random inversion results under different variable distance parameters. From the experimental results, the inversion prediction precision will be higher in this block under the following situations: Expo variation function, horizontal variable distance is 3000 meters and vertical variable distance is 3 meters.