Objective Pain interference and pain behavior are highly relevant outcomes in patients with rheumatoid arthritis (RA). The Patient‐Reported Outcomes Measurement Information System (PROMIS) is a universally applicable set of item banks measuring patient‐reported health, and if applied as computerized adaptive tests (CATs), more efficiently and precisely than current instruments. The objective was to study the psychometric properties of the Dutch‐Flemish PROMIS pain interference (PROMIS‐PI) and the PROMIS pain behavior (PROMIS‐PB) item banks in patients with RA. Methods A total of 2,029 patients with RA completed the full PROMIS‐PI (version 1.1, 40 items), and 1,554 patients completed the full PROMIS‐PB (version 1.1, 39 items). The following psychometric properties were studied: unidimensionality, local dependence, monotonicity and graded response model (GRM) fit, cross‐cultural validity (differential item functioning DIF for language Dutch versus Flemish), other forms of measurement invariance, construct validity, reliability, and floor and ceiling effects. Results The PROMIS‐PI and PROMIS‐PB banks were sufficiently unidimensional (Omega‐hierarchical Omega‐H 0.99, 0.95, and explained common variance 0.95, 0.78, respectively), had negligible local dependence (0.3–1.4% of item pairs), good monotonicity (H 0.75, 0.46), and a good GRM model fit (no misfitting items). Furthermore, both item banks showed good cross‐cultural validity (no DIF for language), measurement invariance (no DIF for age, sex, administration mode, and disease activity), good construct validity (all hypotheses met), high reliability (>0.90 in the range of patients with RA), and an absence of floor and ceiling effects (0% minimum or maximum score, respectively). Conclusion Both PROMIS‐PI and PROMIS‐PB banks showed good psychometric properties in patients with RA and can be used as CATs in research and clinical practice.