The present study aimed to explore the bilingual disadvantage in the course of speech production by comparing the naming performance and the temporal dynamics of object naming in three groups of participants: monolinguals, L1 bilinguals and L2 bilinguals. To determine the origin of the bilingual disadvantage, whether lexical or post-lexical, we manipulated the lexical frequency of the picture names across repetitions. The ERP frequency effect (ERPs associated to high versus low-frequency words) across repetitions, taken as an estimate of lexical processing, was explored before and after articulation was engaged and compared across groups. Additionally, to explore the extent of the bilingual disadvantage in the course of naming, correct responses and errors were compared across groups in the same time-ranges in which lexical frequency was measured. The results before articulation showed earlier frequency effects for monolinguals than bilinguals, while no differences were observed between the two groups of bilinguals. Correctness effects appeared with a similar chronometry as the frequency effects and only for monolinguals. After articulation onset, the three groups of participants showed a larger positivity for low-frequency words and only monolinguals showed frontal error-related activity, supporting that monitoring mechanisms in charge of detecting conflicting responses are influenced by bilingualism. These results suggest a lexical origin of the bilingual disadvantage, with bilinguals being delayed relative to monolinguals in engaging lexical selection. Furthermore, they suggest that self-monitoring might not be implemented in the same way in bilinguals and monolinguals.