In this paper, performance of numeric tabu search algorithm developed by authors on AR (Auto Regressive) system modeling was compared to adaptive methods and artificial neural networks. The adaptive methods are Least Squares Lattice, Double Lattice, Affine Projection algorithm, Least Mean Squares (LMS), Normalized LMS and Recursive Least Squares (RLS). Artificial neural networks trained with eight learning algorithms and classic and numeric tabu search algorithms were the rest of algorithms compared with. The performances of total 16 algorithms were tested on two AR systems having 4th and 6th orders. The numeric tabu search algorithm was found the most successful algorithm on linear AR systems.