Abstract: Multiple-point geostatistics approach uses “training image” to replace variogram and adopt the sequential algorithm to stochastic model，so it combines the advantages of the easy conditioning of pixel-based algorithms and the ability to reproduce object geometry of object-based algorithms.With the analysis of traditional stochastic modeling methods，the paper presents the basic principle and methods of multiple-point geostatistics and SNESIM algorithm，and a case study of offshore reservoir of the Zhujiang Formation of Neogene in a certain block of NanHai West Zhujiangkou basin is made using this technique for the first time.The result indicates that this approach is better than the traditional methods to reproduce the object geometry.Finally，the paper discusses some main problems of this approach，including the training image，object continuity etc.

Key words: multiple-point geostatistics, training image, variation function, SNESIM algorithm, reservoir stochastic modeling, offshore reservoir, longshore bar