We have constructed a reference model to facilitate comparison of serum IGF-I values among children, and thereby to improve the value of IGF-I measurements for diagnosis. The data set consists of serum values measured in 969 samples from 468 healthy children and adolescents (232 males, 236 females; ages, 1.1-18.3 yr). One sample per child was used for the model, each being selected so as to provide sufficient observations for each stage of puberty. The samples not selected were used to validate the reference data. The IGF-I values were log transformed, and multiple regression analysis was used in the model-building process. The best linear model, which converts serum IGF-I concentrations into SD scores and explains 66% of the variation in logIGF-I values, includes the variables of age, gender, and puberty, and takes the interactions among these variables into account. In prepubertal and early pubertal children, the relationship between age and logIGF-I was positive, with greater effect in girls older than 8 yr. In mid-puberty, logIGF-I values were higher in girls than in boys of the same age, up to 16 yr of age. Among boys, the most pronounced positive relationship between age and logIGF-I occurred in mid-puberty, whereas the relationship between age and logIGF-I among girls in mid-puberty is fairly constant. In late puberty, logIGF-I values were higher than earlier in puberty, and there was a negative relationship with age in both boys and girls. Instead of separate models for each combination of puberty and gender, estimating a single regression model permits simultaneous estimation of all explanatory variables and uses all observations in the data set, thereby making it easier to select those variables that have a significant effect on logIGF-I. Our model shows that IGF-I levels are related to age during each stage of puberty. The model also accounts for the fact that serum IGF-I concentrations during puberty are different for boys and girls.