The most commonly used measure of anxiety sensitivity is the 36-item Anxiety Sensitivity Index-Revised (ASI-R). Exploratory factor analyses have produced several different factors structures for the ASI-R, but an acceptable fit using confirmatory factor analytic approaches has only been found for a 21-item version of the instrument. We evaluated the fit of all published factor models for the 36- and 21-item ASI-R, modified the hierarchical model using an approach that does not eliminate items, evaluated the invariance of the modified model across Caucasian and African-American subsamples, and compared the reliability and validity of the 36-item and 21-item versions. The 21-item version of the ASI-R fit a four factor model, as did the 36-item version after several meaningful model modifications. The modified 36-item model was replicable in independent cases and its structural properties were generally invariant across race. Scores from the 36-item version exhibited superior reliability and criterion-related validity.