OBJECTIVES To estimate the minimally important differences (MIDs) for the EQ-5D, HUI2, HUI3, and SF-6D health index scores using health-state transitions described by each instrument's health classification systems as anchors. METHODS We assume that the smallest differences in health states defined by each instrument's multi-attribute health classification (MAHC) systems are associated with important differences in health preferences. Based on this assumption, the MID was defined as the difference in index score between two health states defined by each MAHC system differing in only one health dimension or attribute and by only one functional level. Thus, for each instrument, we enumerated all the theoretically possible pairs of minimally different health states and calculated the differences in index scores for those pairs of health states. RESULTS Based on our definitions, the total number of pairs of minimally different health states is 405 for the EQ-5D, 127,600 for the HUI2, 6,382,800 for the HUI3, and 86,700 for the SF-6D. The mean (standard deviation) MID estimate was 0.040 (0.026) for the EQ-5D (US algorithm), 0.082 (0.032) for the EQ-5D (UK algorithm), 0.045 (0.039) for the HUI2, 0.032 (0.027) for the HUI3, and 0.027 (0.028) for the SF-6D. The effect sizes corresponding to these MID estimates range from 0.19 to 0.28. In general, these MID estimates are quite comparable to those estimated using other anchor-based methods. CONCLUSIONS This new approach to estimating the MIDs of four commonly used preference-based HRQoL index scores provides new and useful information for identifying and interpreting meaningful change (or differences) in scores.