Abstract:The support vector machine (SVM),a new generation machinery learning tech-nology based on statistical theory,has been reported to have better prediction performance of non-liner regression than traditional statistical methods.An SVM regression (SVMR) model for atmospheric pollution prediction is developed according to seven forecast factors, including the daily average pollutant concentration of previous day,daily average wind speed of previous day,etc.Meanwhile,10-fold cross-validation and grid-search methods are ap-plied to find the best parameters of SVMR.The experimental results of Urumqi data show that SVM has the unique advantage of high prediction accuracy and training rate on small-size data sets.It suggests a new model for prediction of atmospheric pollution.