We determined the improvement in baseline length precision and accuracy using new atmospheric delay mapping functions (Ifadis (Ifadis, 1986) and MTT (Herring, 1992a)) by analyzing the NASA Crustal Dynamics Project research and development (R&D) experiments and the International Radio Interferometric Surveying (IRIS) A experiments. These mapping functions reduce baseline length scatter by about 20% below that using the CfA2.2 (Davis et al., 1985) dry and Chao (Chao, 1974) wet mapping functions. With the newer mapping functions, average station vertical scatter inferred from observed length precision (given by length repeatabilities) is 11.4 mm for the 1987–1990 monthly R&D series of experiments and 5.6 mm for the 3-week-long ERDE series. The inferred monthly R&D station vertical scatter is reduced by 2 mm or by 7 mm in a root-sum-square (rss) sense. Length repeatabilities are optimum when observations below a 7–8 ° elevation cutoff are removed from the geodetic solution. Analyses of IRIS-A data from 1984 through 1991 and the monthly R&D experiments both yielded a nonatmospheric unmodeled station vertical error of about 8 mm. In addition, analysis of the IRIS-A experiments revealed systematic effects in the evolution of some baseline length measurements. The length rate of change has an apparent acceleration, and the length evolution has a quasi-annual signature. We show that the origin of these effects is unlikely to be related to atmospheric modeling errors. Rates of change of the transatlantic Westford-Wettzell and Richmond-Wettzell baseline lengths calculated from 1988 through 1991 agree with the NUVEL-1 plate motion model (Argus and Gordon, 1991) to within 1 mm/yr. Short-term (less than 90 days) variations of IRIS-A baseline length measurements contribute more than 90% of the observed scatter about a best fit line, and this short-term scatter has large variations on an annual time scale.