Two probabilist methods of age prediction in children are proposed: they are both based on the radiological presence of erupted teeth or germs. Using an apprenticeship sample of known age and sex, we established several discriminant models (+/- 13, +/- 16, +/- 18 years old). We also evaluated a Bayesian model with the following age groups: < 13, [13-16[, [16-18[, > or = 18 years old, or [X and Y] years old. When applied on a known test sample, Fisher's linear functions presented a success rate greater than 90%, above 13 years threshold, and below 16 and 18 years thresholds, and Bayesian approach, greater than 85%. Therefore, these methods provide an interesting alternative for children age determination that can be applied in biological and forensic anthropology, too.