OBJECTIVES The main objective of this study is to estimate the impact of the level of functional status and disability on health service costs related to rheumatoid arthritis (RA) disease in Hungary. It is straightforward to think that higher disability implies higher costs, where the nature of the relationship is unclear. In order to explore the dependence structure a novel approach is proposed. Instead of fitting trend lines for the cost by regression methods the entire bivariate distribution was modelled.   METHODS Health Assessment Questionnaire’s disability index data were collected for 497 RA patients (with 2594 observations) treated in the Arthritis Center (AC) Buda Hospital of Hospitaller Brothers of St. John from 1st January 2005 to 31st July 2013. The same patients were also found in the database of National Health Insurance Fund Administration (NHIFA) and further parameters as e.g. relevant treatments and health service costs (in- and outpatient) were collected. After merging AC and NHIFA database the 2 dimensional patterns of the HAQ-index measurements versus costs (sum of relevant costs in the following quarter year) become available for bivariate modelling. The ingredients were the empirical distribution of HAQ-index and quarter year cost, and some parametric copula families (elliptical or Archimedean) capturing the (possibly non-linear) dependence structure. The performance of the different model assumptions were compared by goodness-of-fit procedures. RESULTS The fitted bivariate distribution based on the best-performing dependence model are shown in the original “HAQ-index vs. cost” scale. The differences of average costs for low/medium/high HAQ-index values are summarized and the conditional distribution functions of costs are presented, respectively. CONCLUSIONS It has been proved that there is a significant positive dependence between the disability status of RA and health service costs. The dependence cannot be considered as linear but this non-linearity can be tackled easily by using copula methods.