OBJECTIVES: The relationship between severity of lower limb spasticity and quality of life has not been previously characterized. We sought to quantify this relationship among adults with lower limb spasticity from stroke or traumatic brain injury. METHODS: Data on 381 patients from a randomized controlled trial (NCT01249404) and its open-label extension (NCT01251367) were included in our analyses. We fit random intercept linear regression models using the EQ-5D and SF-6D (converted from SF-36) index scores as dependent variables and the Modified Ashworth Scale (MAS) or Physician’s Global Assessment (PGA) as the primary independent variables. All candidate models included age and sex, and tested possible combinations of other covariates using a 10-fold (9 training, 1 prediction) cross-validation procedure. Model selection was based on minimizing the mean squared error (MSE). RESULTS: The four sets of models each converged successfully, with age, sex, prior toxin use, physiotherapy use, other spasticity medication use, and baseline pain level included in all of the final models except the SF-6D and MAS model (physiotherapy was dropped). The MSE of both the EQ-5D models of MAS and PGA was 0.054. The MSE of the SF-6D models was substantially lower at 0.016 and 0.014, respectively. The MAS:SF-6D model produced utility predictions of 0.64, 0.63, 0.63, 0.62, 0.62, and 0.61 for MAS levels 0, 1, 1+, 2, 3, and 4 respectively. The PGA:SF-6D model produced utility predictions of 0.61 (non-response) and 0.63 (response). Most of the utility predictions had overlapping 95% confidence intervals. CONCLUSIONS: The SF-6D models resulted in more accurate predictions of health-related quality of life compared to the EQ-5D models. However, the differences in predicted utility values between the levels of both MAS and PGA do not appear to be meaningfully different for spasticity patients.