In the context of the MAGIC project (Meteorological Applications of GPS Integrated Column Water Vapor Measurements in the Western Mediterranean), zenith tropospheric delays (ZTD) are computed from Global Positioning System (GPS) signals at permanent sites in the western Mediterranean area continuously since November 1998. These time series are compared to the equivalent values estimated by integrating the atmospheric column derived from radiosonde profiles and the HIRLAM numerical weather prediction model in order to determine the error statistics of the data. From a year and a half of data, we show that the difference between radiosonde and GPS ZTD has a standard deviation of 12 mm and a bias of less than 5 mm for most stations. The biases between the GPS ZTD and HIRLAM estimates are smaller, but the standard deviation is greater, usually on the order of 17 mm. The standard deviation of the residuals from both comparisons depends greatly on the humidity which produces an annual signal because of the much higher humidity variability in the summer months.