Abstarct: Wind is one of the most important components in the climate of each region, and its variations can affect phenomena such as dust storm, and the rate of evapotranspiration. Wind behavior prediction due to its very random nature is very challenging and depends on the weather conditions and the regional hydroclimatic and topography factors. In addition to traditional physical and statistical methods, in recent years, some advanced methods baxsed on artificial intelligence have been investigated to achieve high-certainty and accurate predictions. In this study, several models have been developed to predict daily wind speed and direction in a special time of the day. Three types of artificial neural networks, namely NARX, multilxayer perceptron (MLP), and the multilxayer perceptron artificial neural network with Bagging method (bootstrap aggregating) as its data-sampling method were developed to predict one step ahead wind speed. All models were trained by using wind speed data of the Global Forecasting System (GFS) and some observed meteorological parameters (wind speed, air pressure, air temperature, relative humidity and precipitation) collected over a period of approximately two years (October 2014 - July 2016) in Bandar-Mahshahr. baxsed on the results, the multi-laxyer perceptron artificial neural network with Bagging sampling showed better performance compared to other two neural networks. In this research, the effect of different parameters on prediction models has also been investigated. In different input combination, air pressure and air temperature have the most effect on prediction of wind speed in Bandar-Mahshahr.