Due to the depletion of natural petroleum resources in one hand and the increasing trend of demand from industrial sectors in recent years on the other hand, the need for developing new efficient procedures for exploration and production of petroleum resources has attracted more research works in recent years. One way to overcome such difficulties is to improve the currently available multiple point geostatistical techniques known as MPG or MPS in literature which are capable of simulating heterogeneous and complicated petroleum reservoirs with non-linear structures. So far, a number of algorithms have been developed to implement different types of multiple point methods. In current research, the advantages and disadvantages of commonly employed MPS algorithms are evaluated in order to develop new and more efficient MPS algorithms with less pitfalls. In this regard, we have tried to not only maintain the merits of previous methods but to add more advanced and efficient capabilities such as application of pattern recognition and image processing techniques in recovering patterns hidden in the training images. Such techniques could improve the validity and accuracy of realizations produced in the course of simulation procedure. Because of the commercially available and popularity of Zhang’s original Filtersim algorithm and its later developments, we have focused our research work on improving Filtersim algorithm throughout current research. In the first part of current thesis, we have introduced new adaptive filters derived from inferred training images based on principal components analysis. Such approach has not only reduced the dimension of pattern space to the number of filters in filter score space but could recover more realistic patterns analogous to the patterns existing in the training images. The results showed more consistent simulations than those obtained by conventional Filtersim algorithm which uses fixed and predefined gradient filters. Next, a newly developed algorithm based on our improved Filtersim called Random Partitioning Algorithm (RPA) was introduced to reduce both the computing time and improving the continuity of the recovered patterns. In RPA, advantageous aspects of both raster path and random partitioning methods of searching in the simulation procedure simultaneously. In the third part of current research, a number of 10 popular template matching algorithms of pattern recognition method were employed their performance in recovering patterns derived from training images in the course of simulation were compared. Results showed that although none of them was able to outperform others based on evaluation criteria but the BSAD algorithm was the best where the data were more noisy and the Coarse to Fine with Entropy (CFE) method is the best in terms of computing time and memory consumption. In an attempt to test the capability of the newly developed algorithms in tackling real world problems, the real effective porosity data obtained from laboratory core samples and estimated values derived from well log data, all taken from 32 observatory wells were tried. The results showed the effectiveness and capabilities of our algorithms in reproducing complex and folded structure of the reservoir with great accuracies. Filtersim algorithm, Petroleum Reservoir, Multiple point Geostatistics, Adaptive Filters, Geostatistical Simulation, Template Matching Keywords: 1. Filtersim algorithm, 2. Petroleum Reservoir, 3. Multiple point Geostatistics, 4. Adaptive Filters, 5. Geostatistical Simulation, 6. Template Matching