Based on the basic idea that the support vector regression (SVR) can be regarded as a classification problem in the
dual space, a regression algorithm, minimum variance support vector regression (MVSVR), is proposed through constructing a classification problem by using the minimum class variance support vector machines (MCVSVMs). This method inherits the characteristics of the MCVSVMs, and can be transformed into the traditional SVR. The linear and nonlinear cases of the MVSVR are discussed. Experimental results on the artificial and real datasets show the effectiveness of the MVSVR.