Abstract Two recent studies analyzed thyroid cancer incidence in Belarus and Ukraine during the period from 1990 to 2001, for the birth cohort 1968 to 1985, and the related (131)I exposure associated with the Chernobyl accident in 1986. Contradictory age-at-exposure and time-since-exposure effect modifications of the excess relative risk (ERR) were reported. The present study identifies the choice of baseline modeling method as the reason for the conflicting results. Various quality-of-fit criteria favor a parametric baseline model to various categorical baseline models. The model with a parametric baseline results in a decrease of the ERR by a factor of about 0.2 from an age at exposure of 5 years to an age at exposure of 15 years (for a time since exposure of 12 years) and a decrease of the ERR from a time since exposure of 4 years to a time since exposure of 14 years of about 0.25 (for an age at exposure of 10 years). Central ERR estimates (of about 20 at 1 Gy for an age at exposure of 10 years and an attained age of 20 years) and their ratios for females compared to males (about 0.3) turn out to be relatively independent of the modeling. Excess absolute risk estimates are also predicted to be very similar from the different models. Risk models with parametric and categorical baselines were also applied to thyroid cancer incidence among the atomic bomb survivors. For young ages at exposure, the ERR values in the model with a parametric baseline are larger. Both data sets cover the period of 12 to 15 years since exposure. For this period, higher ERR values and a stronger age-at-exposure modification are found for the Chernobyl data set. Based on the results of the study, it is recommended to test parametric and categorical baseline models in risk analyses.