Determination of dry zenith delay, DZD, is important for separating wet zenith delay with high accuracy from Global Positioning System (GPS) measurements. The existing models are usually for the calculation of the zenith hydrostatic delay, ZHD. Because the zenith hydrostatic delay includes the contribution of water vapor and is also not accurate enough, the DZD, approximately replaced by ZHD, will induce ~2 cm error into local GPS applications. In this study we calibrate a popular ZHD model, namely, the Saastamoinen model for local GPS meteorology, using radiosonde data. We test the calibrated model by comparing the predicted delays from the calibrated model against the observed delays from radiosonde data that are not included in the calibration solution. The calibrated model shows 20 mm improvement over the uncalibrated model predictions over the same period. A comparison of the DZDs calculated from GPS measurements over a month period with calibrated and uncalibrated ZHD model predictions shows 15 mm average improvement. ¿ 2000 American Geophysical Union