Radiation exposure at a young age is a strong risk factor for thyroid cancer. We conducted a nested case-control study of 69 thyroid cancer cases and 265 controls from a cohort of 14,054 childhood cancer survivors to evaluate the shape of the radiation dose-response relationship, in particular at high doses, and to assess modification of the radiation effects by patient and treatment characteristics. We considered several types of statistical models to estimate the excess relative risk (ERR), mainly guided by radiobiological models. A two-parameter model with a term linear in dose and a negative exponential in dose squared provided the best parsimonious description with an ERR of 1.3 per gray (95% confidence interval 0.4-4.1) at doses below 6 Gy and a relative decrease in ERR of 0.2% per unit dose squared with increasing dose, that is, decreases in the ERR/Gy of 53% at 20 Gy and 95% at 40 Gy. Further analyses using spline models suggested that the significant nonlinearity at high doses was characterized most appropriately as a true downturn rather than a flattening of the dose-response curve. We found no statistically significant modification of the dose-response relationship by patient characteristics; however, the linear parameter (i.e., the ERR/ Gy at doses less than 6 Gy) did decrease consistently and linearly with increasing age at childhood cancer diagnosis, from 4.45 for 0-1-year-olds to 0.48 for 15-20-year-olds. In summary, we applied models derived from radiobiology to describe the radiation dose-response curve for thyroid cancer in an epidemiological study and found convincing evidence for a downturn in risk at high doses.