OBJECTIVE: To determine variability of arthritis prevalence in 4 US population health surveys. METHODS: We estimated annualized arthritis prevalence in 2011/12, among adults ≥ 20 years, using 2 definition methods, both based on self-report: 1) doctor/health care provider diagnosed arthritis in the Behavioral Risk Factor Surveillance Survey (BRFSS), National Health and Nutrition Examination Survey (NHANES), National Health Interview Survey (NHIS), and Medical Expenditure Panel Survey (MEPS); and 2) three ICD-9-CM based arthritis definitions in MEPS (National Arthritis Data Workgroup Arthritis and Rheumatic Conditions [NADW-AORC], Clinical Classification System [CCS], and Centers for Disease Control and Prevention [CDC]). RESULTS: Diagnosed arthritis prevalence percentages were within 3 percentage points (BRFSS= 26.2% [99% CI=26.0 - 26.4], MEPS= 26.1 [99% CI=25.0-27.2], NHIS=23.5 [99% CI = 22.9-24.1], NHANES=23.0% [99% CI=19.2-26.8]) and ICD-9-CM within 5 (CCS=25.8%; 99% CI=24.6-27.1; CDC=28.3%; 99% CI=27.0-29.6; and NADW=30.7%; 95% CI=29.4-32.1). Range in estimated number (in millions) affected with diagnosed arthritis was 7.8 (BRFSS=58.5 [99% CI=58.1-59.1]; MEPS=59.3 [99% CI=55.6-63.1]; NHANES=51.5 [99% CI=37.2-65.5], and NHIS=52.6 [99% CI=50.9-54.4]) and ICD-9-CM definitions was 11.1 (CCS=58.7 [99% CI=54.5-62.9]; CDC=64.3 [99% CI=59.9-68.6], and NADW=69.9 [99% CI=65.2-74.5]). Most (57% to 70%) reporting diagnosed arthritis also reported ICD-9-CM arthritis; respondents reporting diagnosed arthritis were older than those meeting ICD-9-CM definitions. Proxy response status affected arthritis prevalence differently across surveys. CONCLUSION: Public health practitioners and decision makers are frequently charged with choosing a single number to represent arthritis prevalence in the US population. We encourage them to consider the surveys' purpose, design, measurement methods, and statistical precision when choosing an estimate.