Runoff in the upper Indus in Pakistan is primarily fed by meltwater from snow and ice. Successful modelling of runoff thus depends on knowledge of the energy inputs for melt, and temperature provides a practical index. In this study, spatial and altitudinal variations in air temperature are investigated using correlation and regression analysis. The high levels of seasonal correlation between widely separated stations and with altitude suggest that conditions over a wide surrounding area and up to the freezing level may be inferred with reasonable reliability from climate stations at the valley level. Investigation of concurrent daily rainfall, temperature and runoff in extreme monsoon incursions shows that precipitation is accompanied by a sharp fall in temperature, reduced ablation and, most frequently, a decrease in river flow. Such temperature reductions have practical implications for short term flood forecasting and for design flood estimation.