Hyperspectral image classification is an important research aspect of hyperspectral data analysis. Relevance vector machine (RVM) is widely utilized since it is not restricted to Mercer condition and does not have to set the penalty factor. Due to the high dimension of hyperspectral data, the classification accuracy is severely affected when there are few training samples. Feature reduction is a common method to deal with this phenomenon. However, most of the filter model based feature selection methods can not provide optimal feature selection number. This paper proposes to utilize the statistic estimation characteristic of Monte Carlo random experiments to calculate optimal feature reduction number and conduct hyperspectral image classification with relevance vector machine. Experimental results show the reliability of the feature reduction number calculated by Monte Carlo method. Compared with the classification of original data, there is a significant improvement in the classification accuracy with the feature reduction data.