Speech production is one of the most fundamental activities of humans. A core cognitive operation involved in this skill is the retrieval of words from long-term memory, that is, from the mental lexicon. In this article, we establish the time course of lexical access by recording the brain electrical activity of participants while they named pictures aloud. By manipulating the ordinal position of pictures belonging to the same semantic categories, the cumulative semantic interference effect, we were able to measure the exact time at which lexical access takes place. We found significant correlations between naming latencies, ordinal position of pictures , and event-related potential mean amplitudes starting 200 ms after picture presentation and lasting for 180 ms. The study reveals that the brain engages extremely fast in the retrieval of words one wishes to utter and offers a clear time frame of how long it takes for the competitive process of activating and selecting words in the course of speech to be resolved. electrophysiology lexical access speech production W ord selection is a crucial step in speech production. Considering that the average lexicon contains 50,000 lexical entries and that an average speaker utters approximately three words per second, the process of lexical retrieval needs to proceed at high speed and with great accuracy. Failures of this process result in speech errors or anomia, which limit communication , as acutely demonstrated in production aphasia, for instance. Although our understanding of how speakers retrieve words from the lexicon has considerably increased in recent years (1-4), the neural implementation of this process remains poorly understood. In particular, insights regarding the time course of word retrieval in speech production are sparse, and most of the chronometric evidence available is derived from event-related potential (ERP) studies relying on button-press responses rather than in actual overt speech production (5-9). This strategy was adopted because EEG is highly susceptible to mouth movements that could possibly mask the cognitive components of interest. However, at least one EEG study and several MEG studies have shown that artifact-free brain responses can be measured up to at least 400 ms after picture onset (10-13), and a few recent ERP studies demonstrated that classical ERP components can be replicated during overt picture naming (14-17). Although these latter studies reveal the validity of ERPs for studying overt naming, they have not directly investigated the issue of the time course of lexical selection, but rather other aspects of word production (e.g., morphological processing, bilingual language control, etc.). It is the goal of the present study to identify the time course of word selection during overt naming, capitalizing on the fine temporal resolution of ERPs. In this study, we directly measure the time course of word retrieval during overt naming. Such temporal information is invaluable for understanding brain mechanisms underlying speech production and to elucidate the causes of speech failure. Additionally, this approach may offer a framework to constrain theoretical models of language production. With this aim in mind, we chose to exploit the cumulative semantic interference effect (CSIE) (18) as a proxy for lexical selection in picture naming. In this paradigm, participants name a set of pictures from intermixed semantic categories (e.g., turtle, hammer, tree, crocodile, bus, axe, snake, etc.). Interestingly, naming latencies for a given picture (e.g., snake) increases linearly with its ordinal position within the semantic category. That is, the naming latency for a given item correlates positively, linearly, and highly with the number of items from the same semantic category that have been named previously (e.g., turtle, crocodile). It has been debated whether this striking effect reflects competition at a semantic or a lexical level (18-23). According to the semantic account, the effect arises at the level at which semantic representations are being selected for production. The idea is that semantically related conceptual representations compete with one another, such that the reactivation of previously selected representations hinders the speed with which the semantic representation corresponding to the target picture is singled out and selected for further processing (22, 23). According to the lexical interpretation, the interference between semantic competitors is assumed to reflect the difficulty in selecting a specific lexical item produced by the reactivation of lexical items of the same semantic category that have been produced previously: The more items of the same semantic category are encountered, the higher the amount of competition when selecting another item (the target one) from the same semantic category. Although there is still debate about the specific origin of the effect, several fMRI and MEG studies have consistently shown increased activation during semantic competition in the middle temporal gyrus and the left inferior frontal gyrus (12, 24-28), two regions putatively associated with lexical access and selection (29-31). Knowledge of the timing of ERP effects elicited by the CSIE is likely to provide key temporal information on lexical selection during picture naming. More specifically, we predict that there should be a time window in which the ERPs correlate positively with the ordinal position of the target within the series of pictures belonging to the same semantic category. To the extent that the CSIE reveals the processes engaged during word selection, such a window can be considered as the time interval in which word retrieval takes place. There are two independent studies that allow for a precise prediction about the time at which word retrieval from the lexicon occurs. First, according to the meta-analysis conducted by Indefrey and Levelt (29), lexical selection will unfold between