This chapter began by presenting a recent connectionist model of lexical access in production, the semantic-phonological model. This model emphasized the two-step nature of access, but allowed each access step to be carried out by interactive activation. Although that model provided a good account of the basic error categories associated with the access of single words by normal and aphasic speakers, it suﬀered from a number of limitations, particularly with respect to its mechanisms for phonological processes. Models that make use of time-varying control signals have promise in addressing these limitations, by providing for eﬀects of linguistic structure and temporal sequence on phonological encoding. Ultimately, though, production models need to take greater advantage of connectionist learning mechanisms and address data that are informative about how the language production system changes with experience.