The paper discusses the impact of atmospheric modeling errors on the estimation of baseline length in very long baseline interferometry (VLBI) data. Systematic errors in prior estimates of baseline length, on the order of 5 cm for baselines around 8000 km, are attributed primarily to mismodeled atmospheric delay, specifically the troposphere and mesosphere. The authors present observational evidence for these errors and develop a new "mapping" function to account for the elevation angle dependence of atmospheric delay. This new mapping function, derived from ray tracing through model atmospheres, introduces errors into baseline length estimates of ≤ 1 cm, significantly reducing the systematic errors compared to the previously used models. The paper also includes a detailed derivation of the zenith delay formula and discusses the accuracy of the hydrostatic (dry) delay model. The new mapping function, dubbed CFA-2.2, is validated through an elevation angle cutoff test, showing improved accuracy in baseline length estimates.The paper discusses the impact of atmospheric modeling errors on the estimation of baseline length in very long baseline interferometry (VLBI) data. Systematic errors in prior estimates of baseline length, on the order of 5 cm for baselines around 8000 km, are attributed primarily to mismodeled atmospheric delay, specifically the troposphere and mesosphere. The authors present observational evidence for these errors and develop a new "mapping" function to account for the elevation angle dependence of atmospheric delay. This new mapping function, derived from ray tracing through model atmospheres, introduces errors into baseline length estimates of ≤ 1 cm, significantly reducing the systematic errors compared to the previously used models. The paper also includes a detailed derivation of the zenith delay formula and discusses the accuracy of the hydrostatic (dry) delay model. The new mapping function, dubbed CFA-2.2, is validated through an elevation angle cutoff test, showing improved accuracy in baseline length estimates.