Log-linear, exponential and fractional relations for estimating seasonal\nsnowmelt from early-spring snow accumulation in the Indus and Kabul\nriver basins in the western Himalayas are developed with a view to\nimprove the prediction given by bivariate linear regression models\nearlier developed by the senior author in collaboration with others.\nThis study shows that although the transformed data may improve the\nabove prediction, they fail to satisfy the condition of nonlinearity; a\nproperty that must be borne in mind before recommending any nonlinear\nregression model. Any further improvement in the prediction of seasonal\nflow volume from basin snow cover area, therefore, has to come from\nwithin the domain of linear regression models only or from improvements\nin the original input data.