The purpose of this study is to present a simple but explicit model of on-the-job training which may enable us to separate and identify various types of vintage effects. An attempt is made to apply the model to the data on the earnings of American scientists in the period 1960-1970