Many studies have shown that isotope data are valuable for hydrological model calibration. Recent developments have made isotope analyses more accessible but event sampling still involves significant time and financial costs. Therefore, it is worth to study how many isotope samples are needed for hydrological model calibration and what the most informative sampling times are. In this study, we used synthetic data to investigate how systematic errors in the precipitation, streamflow and the isotopic composition of precipitation affect the information content of stream isotope samples for model calibration. The results show that model performance improves significantly when two or three isotope samples are used for calibration and that the most informative samples are taken on the falling limb. However, when there are errors in the rainfall isotopic composition, rising limb samples are more informative. Data errors caused the most informative samples to be more clustered and to occur earlier in the event compared to error free data. These results provide guidance on when to sample events for model calibration and thus help to reduce the cost and effort in obtaining useful data for model calibration.