In K-NN model, cross validation is a technique to estimate the optimum number of nearest neighbors with minimum cross validation error. However the trial and error procedure followed in this technique makes it very rigorous and time consuming. Therefore, to overcome the disadvantages associated in the existing cross validation technique, a new cross validation technique has been proposed in this study which is based on independent variables of the system. To predict daily discharge of five monsoon months of Tikarpada gauging station of Mahanadi basin, K-NN models have been applied with the proposed cross validation technique. In this technique, based on good correlation between daily discharge of Tikarpada and that of other gauging stations of Mahanadi basin, independent variable of the system were selected and number of nearest neighbors of the K-NN models were fixed according to number of independent variables. High discharge prediction performance of the K-NN models indicated high efficiency of the proposed cross validation technique. Therefore, the proposed cross validation technique can be adopted for developing K-NN models for other gauging stations of Mahanadi basin and also for other basins.