Geodetic monitoring of subduction of the Nazca and Cocos plates is a goal of the CASA (Central and South America) Global Positioning System (GPS) experiments, and requires measurement of intersite distances (baselines) in excess of 500 km. The major error source in these measurements is the uncertainty in the position of the GPS satellites at the time of observation. A key aspect of the first CASA experiment, CASA UNO, was the initiation of a global network of tracking stations to minimize these errors. We studied the effect of using various subsets of this global tracking network on long (>100 km) baseline estimates in the CASA region. Best results were obtained with a global tracking network consisting of three U.S. fiducial stations, two sites in the southwest Pacific and two sites in Europe. Relative to smaller subsets, this global network improved baseline repeatability, resolution of carrier phase cycle ambiguities, and formal errors of the orbit estimates. Describing baseline repeatability for horizontal components as &sgr;=(a2+b2L2)1/2 where L is baseline length, we obtained a=4 and 9 mm and b=2.8¿10-8 and 2.3¿10-8 for north and east components, respectively, on CASA baselines up to 1000 km in length with this global network. ¿ American Geophysical Union 1990