The purpose of this research is investigating the usefulness of variables reduction and support vector regression in predicting stock returns of companies listed on Tehran Stock Exchange (TSE). In this regard, through reviewing literature, 52 predictive features were specified as the initial features (variables) based on the popularity in the literature and the availability of the necessary data. By using correlation-based variables selection method and factor analysis variables extraction method, optimal variables (factors) are selected or extracted from initial variables. Subsequently, the stock returns of 101 firms listed on TSE from 2004 to 2013 were predicted utilizing nonlinear methods (support vector regression and artificial neural networks) and linear regression. The research results indicate that support vector regression outperforms the two other prediction methods and both nonlinear methods outperform the linear regression. Furthermore, the results confirmed the usefulness of variables reduction methods and existence of significant difference between usefulness amount of the correlation-based and factor analysis method