Analysis of very long baseline interferometry data indicates that systematic errors in prior estimates of baseline length, of order 5 cm for ∼ 8000-km baselines, were due primarily to mismodeling of the electrical path length of the troposphere and mesosphere (“atmospheric delay”). Here we discuss observational evidence for the existence of such errors in the previously used models for the atmospheric delay and develop a new “mapping” function for the elevation angle dependence of this delay. The delay predicted by this new mapping function differs from ray trace results by less than ∼ 5 mm, at all elevations down to 5° elevation, and introduces errors into the estimates of baseline length of ≲ 1 cm, for the multistation intercontinental experiment analyzed here.