An and Owen (2001) introduced quasi-regression for approximating an unknown function in high dimensions. This approach has very high computational efficiency, particularly when samples size is very large. In this paper, we construct many fitting functions of an unknown function based on quasi-regression, and take the best one from them as the resulting approximation of the unknown function. We show the fitting function obtained has less residual sum of squares than original quasi -regression. An example is given for illustrations at the end of this paper.