Within the framework of item response theory (IRT), we developed a German version of an item bank, as well as a software application that can be employed to measure anxiety by means of a computer adaptive test (CAT). A sample of n = 2348 psychiatric and psychosomatic patients answered a set of up to 13 standardized questionnaires. 81 items drawn from these questionnaires were considered pertinent to the anxiety construct. Various tests were conducted to ensure the suitability of these items for an IRT-based assessment. After these tests, 50 items remained in the item bank and were calibrated using the Generalized Partial Credit Model. Simulation studies conducted on an independent sample of n = 1528 respondents indicate that 6–8 items suffice to measure the latent trait with high precision (standard error ≤ 0.32). CAT scores correlated highly with scores estimated from all available items (r = .97) and scale scores of the State Trait Anxiety Inventory (STAI, state scale, r = .93). Within a routine clinical setting, 102 in-patients answered the Anxiety-CAT along with a number of established anxiety questionnaires. The correlation between the Anxiety-CAT and the STAI state scale was still high (r = .60), but lower than the correlations found in the simulation studies. The Anxiety-CAT was able to differentiate between mental health disorders in a similar manner as established questionnaires. These results suggest that the Anxiety-CAT does indeed exhibit the advantages expected from theory, but the results of further studies are needed in order to judge its full potential for research and clinical practice.