OBJECTIVES: A protocol for the valuation of SF-6D health states began with telling the participants that the “all-worst” health state (645655) was the worst among all health states to be considered. Respondents might decide if “all-worst” was worse or better than dead. This secondary analysis aimed to evaluate whether this practice would have an impact on the valuation results. METHODS: This was a population-based valuation study of the SF-6D, involving totally 1020 participants in Singapore. The SF-6D health states were valued using a visual analogue scale (full health=100 points). This analysis focused on the 73 participants who valued the all-worst health state and at least one health state that was only one step better than the all-worst state in one or two of the six dimensions of the SF-6D (e.g. 645555 and 545654). We call these the “near all-worst” health states. We estimated the label effect (if any) by comparing the value assigned to the all-worst versus the near all-worst health states using graphical means and regression analysis.  RESULTS: A total of 56/73 participants considered the all-worst state worse than death. Among them, the all-worst health state was valued significantly lower than the near all-worst health states (30 points; P<0.001), even after adjustment for the difference attributable to the one step difference in the six dimensions. Among the 17/53 participants who considered the all-worst state better than death, the valuation result was as expected according to the differences in the six dimensions. CONCLUSIONS: The procedure to tell participants that one of the states was “all-worst” had a labeling effect, but not every respondent was affected.