Abstarct: Bridges are one of the important river constructions and local scour at bridge piers is one of the most important factors of their destruction. Lack of attention to hydraulic factors is the destruction cause of many bridges all over the world. Therefore, accurate estimation of local scour around the piers should be taken into account for a good and assured design of bridges. Many researchers has tried to estimate the maximum depth of the scour hole and some empirical relations has been proposed for it. The low accuracy rate of these relations and complex essence of the problem has made the alternative solutions to be considered. Machine Learning techniques are one of these alternatives. Artificial neural network, as one of the most popular machine learning methods, has been extensively used to estimate piers local scour depth. But weighted radial bias function (WRBF) and ensemble methods has not been tried yet. In this research, WRBF and ensemble method’s performance are evaluated and compared to radial bias function (RBF) and empirical methods. The results show that ensemble method with MLP baxse classifiers, outperforms other methods and has more accurate results. The results also indicates the machine learning methods are more accurate than empirical relations. Flow depth, flux rate, the width of the bridge pier, bed particles diameter and critical speed are used as input parameters and longtime pier scour depth is the only output of the neural network. It is shown in the previous researches that frolhich and shen relations underestimate the pier scour depth, Richardson relation has relatively good performance and jin-fisher, melvil-saterland and Larsen and tachrelations overestimation it.