Fundamental to using utilities for future health states for clinical decision analysis is the assumption that the utilities are stable. The purpose of this study was to determine if cancer patients' prior evaluations of a treatment-induced health state remain stable when they later experience that state. Before a four-week course of radiation therapy, 66 laryngeal cancer patients evaluated three common treatment-induced outcome scenarios (mild/moderate/severe) using rating scales and the time trade-off, and provided self-assessments of voice symptoms, voice function, and general health. Evaluations and self-assessments were repeated at the end of therapy, and respondents indicated which outcome scenario described their actual end-of-therapy state. Twenty-four patients identified the mild scenario as their outcome state, 36 the moderate, and 6 the severe. No across-time differences in evaluations were detected, although significant (P < .0001) downward shifts in scores for voice symptoms (t = 5.67), voice function (t = 5.10), and general health (t = 6.69) indicated that actual change in health status had occurred. These results imply that patients' evaluations of imagined short-term health states encountered during radiation therapy for laryngeal cancer remain consistent when those states are experienced at a later time. The design of this project provides a model for determining if this consistency appears in other clinical situations.