OBJECTIVES: Despite EQ-5D’s long establishment, valuation methods have only in recent times been subjected to fresh consideration, mainly driven by the introduction of EQ-5D-5L. However, methods in relation to modelling of health states not directly valued have undergone less change.  This work uses the Irish 3L data set to test a number of alternative methods for the 3L sets.  METHODS: Existing methods employed for valuation studies include fitting logistic regression models to the data.  The addition of interaction terms to capture any extreme health problems (N3), moderate levels (D2), and others, has predominated. Using TTO, a respondent may trade off at some unknown point between two given figures.  We have applied an interval regression model to ensure that the results take into account  such uncertainty. Further we have employed latent variable models to identify subgroups of respondents within the dataset. A reference group can then be identified, rather than excluding respondents according to arbitrary decision rules. RESULTS: A latent variable approach was found to accommodate heterogeneity in the respondent cohort, providing better sensitivity compared to medians (which might otherwise be expected to fulfil a similar role). Interval regression (both log-normalised and otherwise) appears to have only had a small impact on the subsequently derived quality of life in each health state, though it may be considered a more accurate result. The log-normalised interval regression approach also reduced the effect of extreme WTD scores. CONCLUSIONS: We have explored alternative statistical techniques for tackling some of the challenges associated with TTO data. The use of simple regression analysis may not necessarily be the most accurate reflection of population preferences. Techniques such as interval regression and latent variable models should be further investigated in future.