Infants can visually detect changes in numerosity, which suggests that a (non-symbolic) numerosity system is already present early in life. This non-symbolic system is hypothesized to serve as the basis for the later acquired symbolic system. Little is known about the processes underlying the transition from the non-symbolic to symbolic code. In the current study we investigated the development of automatization of symbolic number processing in children from second (6.0 years) and fourth grade (8.0 years) and adults using a symbolic and non-symbolic size congruency task and event-related potentials (ERPs) as a measure. The comparison between symbolic and non-symbolic size congruency effects (SCEs) allowed us to disentangle processes necessary to perform the task from processes specific to numerosity notation. In contrast to previous studies, second graders already revealed a behavioral symbolic SCE similar to that of adults. In addition, the behavioral SCE increased for symbolic and decreased for non-symbolic notation with increasing age. For all age groups, the ERP data showed that the two magnitudes interfered at a level before selective activation of the response system, for both notations. However, only for the second graders distinct processes were recruited to perform the symbolic size comparison task. This shift in recruited processes for the symbolic task only might reflect the functional specialization of the parietal cortex.