The distribution of reservoir is related with the location. When forecasting the reservoir, not only the geological structure and statistical information is investigated, but also the place must be considered to reflect the variability of the reservoir sedimentary pattern, the so called non stationary characteristics. A pattern based multiple point geostatistics is proposed in the paper. The distance function is introduced to characterize the relationship between the sedimentary pattern and the deposit place, the whole substitution, structured random path and multi grid strategy are selected to reproduce the sediment patterns. A synthesized 3 D training image is construced by the constrain of Poyang lake in China and the comparison test is executed. The new designed method is better reproduced the non stationary characteristics than the traditional one, and is more suitable for the simulation of delta front deposits.The research enriches the theory of reservoir stochastic modeling and provides a new choice in real reservoir modeling.