OBJECTIVES: To date, no model has explained the relationship between the NEI-VFQ and health state utilities such as those measured by the EQ-5D. Ordinary Least Square (OLS) is most commonly used to identify association but it may not be appropriate in low vision. In this study, we evaluate different model specifications to identify which better predicts the relative importance of the NEI-VFQ 25 dimensions on the EQ-5D index. METHODS: We compare OLS and Tobit approach using cross-sectional data (n = 155) at screening from a phase I/II clinical trial in patients with neovascular age-related macular degeneration (NV-AMD). We validate the models using a split-sample technique and calculate each model's mean predicted error and standard error. Correlations between the predicted EQ-5D index values derived from the NEI-VFQ 25 dimensions and the observed EQ-5D index score are compared across models. RESULTS: Mean prediction error from the Tobit model is lower than the OLS approach (26 vs. 39 percent). The standard errors of prediction of the Tobit and OLS models are 0.0263 and 0.0234, respectively. The predicted EQ-5D index value from the Tobit model provides better correlation with the observed EQ-5D index score compared to the OLS approach [Pearson Correlation Coefficients are 0.57 and 0.47, respectively]. CONCLUSION: In this situation, the Tobit model provides better predictive accuracy than OLS for explaining the relationship between the EQ-5D index and the NEI-VFQ 25. Tobit produces consistent estimates of the relationship between the EQ-5D index and the dimensions of the NEI-VFQ 25 by accounting for censoring and ceiling effect problems. Although it is sensitive to model misspecification, adjusting for heteroscedasticity nevertheless allows it to perform better than OLS. Verification of these results using the model and a second dataset is warranted.