Utilities are numeric measurements that reflect an individual's beliefs abo
ut the desirableness of a health condition, willingness to take risks to ga
in health benefits, and preferences for time. This report discusses the app
roaches to assess and compare the validity of methods used to assign utilit
ies for cost-utility analysis. Threats to validity include construct underr
epresentation and construct-irrelevant variance. Construct underrepresentat
ion occurs when a stimulus presented to a judge fails to fully represent th
e depth and complexity of information required in actual judgments. Constru
ct-irrelevant variation occurs when factors irrelevant to preferences influ
ence measurements of utilities. Among several factors that cause construct-
irrelevant variation are cognitive abilities, numeracy skills, emotions and
prejudices, and the elicitation procedure. Commonly used elicitation metho
ds (visual-analog scales, time tradeoff, and standard gamble) capture diffe
rent facets of utilities (desirableness of states, time preferences, and ri
sk attitude) to different degrees. The validity of an elicitation protocol
depends (1) on the degree to which its scaling method captures the relevant
facets of utility and (2) on the degree to which measurements are influenc
ed by construct-irrelevant variation. Discrete-state health index models pr
ovide an alternative to direct elicitation of utilities and work by attachi
ng fixed preference weights to observable health states. The creation of di
screte-state models with current technologies requires the adoption of stro
ng assumptions about the scaling properties of utilities. Future research m
ust refine methods of eliciting utilities and identify sources of construct
-irrelevant variability that reduce the validity of utility assessments. Be
cause of the impact of variation in techniques on measurements, we do not r
ecommend the combination of utilities elicited with different protocols in
cost-utility analysis and do not recommend the display of cost-utility rati
os from different studies in comparison or "league" tables.