A comparative evaluation of several rain attenuation prediction models claiming global applicability was performed using data from 126 propagation paths in the CCIR data bank and 76 paths in the Thayer School data bank. The best predictive model relative to the available observations was determined, and the strengths of the different modeling techniques were evaluated. The path-to-path deviations between measurements and model predictions were found to obey a lognormal distribution. This result was obtained for the five prediction models subject to analysis: the Global and revised two-component models by Crane, the revised CCIR model, a modified version of the lognormal model of Morita and Higuti, and a revised version of the “simple” model of Stutzman and Dishman. The root-mean-square differences between measurement and model prediction were model dependent. Overall, the revised CCIR model performed best. For applications in North America the Global model performed best, for Europe the revised CCIR model performed best, and for Asia the revised CCIR model was best. The models were tested in combination with modeled or measured rain rate distributions. The Global rain rate model produced better results than the CCIR rain rate model. When long-term rain rate measurements were available, the best results were obtained by using the measurements.