Geodetic baseline measurements using the Global Positioning System (GPS) were acquired in the Gulf of California between Loreto and Cabo San Lucas in Baja California and Mazatlan on the mainland of Mexico. Tropospheric water vapor content was high during the experiment, typically yielding wet path delays in excess of 20 cm at zenith. Surface meteorological (SM) and water vapor radiometer (WVR) measurements were recorded at each site, providing independent means of calibrating the GPS signal for the wet tropospheric path delay. Residual wet delays at zenith are estimated as constants and also as first-order exponentially correlated stochastic processes. In addition, the entire wet zenith delay is estimated stochastically without prior calibration. The results of these approaches are compared in terms of day-to-day baseline repeatability and other system performance discriminants. Calibration with WVR data yields the best repeatabilities, of the order of 1–7.5 parts in 108 in the horizontal components of 350- to 650-km baselines with carrier phase data. Further improvement in these results occurs if combined carrier phase and pseudorange data are used. WVR data are important for direct characterization of the wet tropospheric path delays in humid regions. SM measurements, if used with a simple atmospheric model and estimation of residual zenith delays as constants, can introduce significant errors in baseline solutions. However, SM calibration and stochastic estimation of residual zenith wet delays may be adequate for precise estimation of GPS baselines, with a deterioration in repeatability of less than 1–2 cm compared to WVR calibration. Stochastic estimation of the entire zenith wet delay yields comparable repeatabilities, particularly if both carrier phase and pseudorange data are used. Similar analyses of the Owens Valley Radio Observatory-Mojave baseline in California, where zenith wet delays are factors of 3–5 less, show no significant differences among the various tropospheric calibrations.