Predictions in ungauged basins have been a major challenge in hydrologic sciences, and there is still much work needed to achieve robust and reliable predictions for such basins. Here, we propose and test a novel approach for predicting runoff from poorly gauged basins using a minimum complex model calibrated with isotope data alone (i.e., without observed discharge data). The model is composed of two water-stores (soil water and groundwater) and considers their connectivity to runoff in terms of both water and isotope budgets. In a meso-scale basin in which riverbed deformations frequently occur, making automatic observation of river discharge difficult, we measured hydrogen and oxygen isotope composition (δ2H and δ18O) of precipitation and river water twice-weekly for one year. Runoff predicted by the model agreed well with that observed monthly or bimonthly. Monte Carlo simulation revealed a strong coherence between model performance in isotope simulation and runoff prediction, demonstrating that the use of isotopes as dynamic proxies of calibration targets helps reliably constrain model parameters. Our results indicate that this approach can serve as a powerful tool for prediction of runoff hydrographs, particularly for basins in which the stage-discharge relationship is highly variable.