OBJECTIVES: The EQ-5D is the most used questionnaire to obtain utilities. Because of its index ceiling effect, Tobit and CLAD-CQR (Censored Least Absolute Deviations – Censored Quantile Regression) models are often used instead of Ordinal Least Squares for its modelization. CLAD models are usually presented as a robust alternative to the Tobit misspecification problems in presence of heteroskedasticity. The aim of this study is to evaluate the suitability of the CLAD-CQR models for the estimation of utilities and QALYs (Quality-Adjusted Life Years), to evaluate the impact of different chronic conditions in Catalan population’s health. METHODS: The EQ-5D was interview-administered in the 2006 Catalan Health Interview Survey, a cross-sectional study of a representative sample of the non-institutionalised general population (n=15,926). As well as CLAD-CQR, Tobit model was adjusted to assess utility losses associated with 15 chronic conditions. Goodness of fit was assessed by cross-validation (Mean Square Error –MSE-, Mean Absolute Deviations –MAD-). RESULTS: As ceiling effect was over 50%, CQR models on the 30 percentile had to be applied instead of CLAD models on the median. CQR showed a slightly worse fit to data (MSE=0.084, MAD=0.246) than Tobit (MSE=0.068, MAD=0.218). The impact of the different chronic conditions measured in QALYs obtained from the CQR was on average around 70% larger than the ones of the Tobit model. CONCLUSIONS: Tobit and CLAD-CQR model latent quality of life, not anchored in 0=death and 1=perfect health. While Tobit allows an estimation of observed utilities and marginal effects, interpretable as QALYs, CQR do not in the case of right-censoring. This leads to overestimation of effects and makes CLAD-CQR models inappropriate for obtaining QALYs, just as the untransformed Tobit’s latent variable. Moreover, the suitability of modelling percentiles differing from the median should be discussed.