OBJECTIVES: To review relevant Rheumatoid Arthritis (RA) economic models for biologics; and identify potential model limitations.  METHODS: Search targeted economic evaluations of RA biologics since 2000 using Medline, EMBASE, and Cochrane databases.  Articles were subjected to a two level review process before data abstraction. RESULTS: Twenty-six economic evaluations were published assessing costs and outcomes associated with RA biologics.  Most models used a payer perspective. Two methotrexate (MTX)-naïve patient models were cost utility analyses (CUA); one was a patient simulation model and one a decision analytic model. Of seventeen models for disease modifying anti-rheumatic drug (DMARD)/MTX failure populations, sixteen were CUAs and one was a cost-effectiveness (CE) model based on cost/ACR improvement achieved; model structures included patient level simulation, Markov, and decision analytic models.  Of seven models identified for anti-TNF inhibitor failure populations, five were CUAs and two were CE models where CE was defined by both cost/remission and cost of achieving low disease activity (Disease Activity Score (DAS)-28 ≤3.2); six models employed a simulation structure and one a Markov structure. Results varied widely across studies due to heterogeneity in the time horizon, perspectives, year of costs and comparators.  Model Incremental cost effectiveness ratios (ICERs) ranged from $4,849 (2007$)- $47,157 (2007$) per QALY for MTX-naïve, $14,518 (1998$) -$498,420 (2005$) per QALY for DMARD/MTX failure, and $12,869 (2006$)-$76,363 (2008$) per QALY for TNF-failure.  Key limitations included limited availability of treatment data over long time horizons, and use of Health Assessment Questionnaire (HAQ) as primary outcome and as determinant of utility.  CONCLUSIONS: We recommend future modeling efforts evaluate the use of direct utilities versus mapping; advantages of CUA versus CE and simulation approach using patient level data; benefits of  longer time horizon; and inclusion of both health related quality of life assessment such as HAQ and disease activity such as DAS-28 as model inputs.