This paper presents systematically the principle and methods of multiple-point geostatistics, and a case study of stochastic modeling is made taking the fluvial reservoir of the Minghuazhen Formation of Neocene in a certain block of Bohai Bay Basin in China as an example. Multiple-point geostatistics is a promising discipline in stochastic reservoir modeling. This approach combines the easy conditioning of pixel-based algorithms with the ability to reproduce object geometry of object-based techniques. It overcomes the drawbacks that traditional variogram-based two-point geostatistics can not express joint variability of more than two locations at a time and reproduce crisp geometries. Through the theory and case analysis some present problems and further study suggestions are discussed such as the stationarity of training images, object continuity, integration of soft information.