The article presents a comprehensive theory of lexical access in speech production, which is a complex but fast and accurate process. The theory is staged and feed-forward, involving conceptual preparation, lexical selection, morphological and phonological encoding, phonetic encoding, and articulation. The core of the theory, from lexical selection to phonetic encoding, is captured in the computational model WEAVER ++. The theory and model have been developed through reaction time experiments, particularly in picture naming paradigms, to account for real-time processing in normal word production. The model can handle speech errors and provides new insights into the cerebral organization of speech production through high-temporal-resolution imaging. The article also discusses the ontogenetic development of word production, from infants' initial babbles to the acquisition of complex sentences. It highlights the duality of the word production system, which involves both conceptual and articulatory motor systems. The theory is based on a feedforward activation-spreading network, with three strata: conceptual, lemma, and form. The network is designed to solve binding problems and account for speech errors. The theory emphasizes the importance of self-monitoring and the role of internal speech in monitoring overt speech output. The article concludes by outlining the general design properties of the theory, including the network structure, competition but no inhibition, binding mechanisms, and relations to the perceptual network. It also discusses the application of Ockham's razor in the theory and model development, emphasizing minimal assumptions to ensure empirical productivity.The article presents a comprehensive theory of lexical access in speech production, which is a complex but fast and accurate process. The theory is staged and feed-forward, involving conceptual preparation, lexical selection, morphological and phonological encoding, phonetic encoding, and articulation. The core of the theory, from lexical selection to phonetic encoding, is captured in the computational model WEAVER ++. The theory and model have been developed through reaction time experiments, particularly in picture naming paradigms, to account for real-time processing in normal word production. The model can handle speech errors and provides new insights into the cerebral organization of speech production through high-temporal-resolution imaging. The article also discusses the ontogenetic development of word production, from infants' initial babbles to the acquisition of complex sentences. It highlights the duality of the word production system, which involves both conceptual and articulatory motor systems. The theory is based on a feedforward activation-spreading network, with three strata: conceptual, lemma, and form. The network is designed to solve binding problems and account for speech errors. The theory emphasizes the importance of self-monitoring and the role of internal speech in monitoring overt speech output. The article concludes by outlining the general design properties of the theory, including the network structure, competition but no inhibition, binding mechanisms, and relations to the perceptual network. It also discusses the application of Ockham's razor in the theory and model development, emphasizing minimal assumptions to ensure empirical productivity.