Abstarct: Investigating of underground discontinuities has an important role in miming activities. Model of these structures can be worth information in various mine’s activity such as Management extraction and Production Planning. The geostatistics introduce many tools for this issue. These tools are baxsed on multiple point geostatistics science and are interested in by who are working in mines and petroleum industry. One of these multiple point geostatistics tools called direct sampling that is used in this thesis to model structure of diyks in sungun copper deposit as one of important copper mine in Iran. Studding of diyks structure of the mine shows these diyks have unstationarity nature so study of such structures must be done in unstationarity geostatistics field. Generating the training image for these structures is very difficult and to generate more stationary training image we need to model patterns of training image. To simulate structure of diyks some scxripts created by Matlab software and sensitivity analysis of these scxripts shows the algorithm is more sensitive to some parameters such as threshold and size of data event and is less sensitive to another one such as fraction and count of hard data. The statistics (single point, two point and connectivity) shows the simulations of diyks successfully done. Also trend reproduction of diyks done correctly. The results are showed on simulated images. Although the connectivity of diyk’s structure on simulated images seem to be less connected than training image, nevertheless the connectivity plot shows there is no considerable difference between both of them. A comparing between Direct Sampling and SNESIM shows Direct Sampling is more able to reproduce diyk’s structures and the statistics confirm this conclusion.