Modeling of tight gas reservoir in less well area on the sea is one of problems which are eager to overcome. Aiming at solving high heterogeneity of tight sandstone gas reservoirs in braided river deposits, it is difficult to accurately characterize reservoir's distribution of "sweet point" and properties using conventional geological modeling method. In this paper, using "sweet point" controlling, multiplepoint geostatistics and probabilistic volume constraint on thick layer reservoirs in the East China Sea, we have innovatively modeled using "sweet point" in a form of facies and generated training image in different ways considering different kinds of "sweet point" and simulated facies based on Poisson's ratio volume and "sweet point" probabilistic volume as a second constraint. The results show that simulating "sweet point" in the form of facies by using multiplepoint can overcome limitation of single facies and characterize the spatial structure of "sweet point". The result is in line with geological knowledge by using different training images in different members. The prediction accuracy of distribution of "sweet point" has been improved under the constraint of probabilistic volume.