The present study builds upon findings from event-related potential (ERP) studies of single word production in order to shed light onto the mechanisms underlying the production of dual-word adjective-noun phrases (NPs). In a first experiment, we tested for potential differences elicited by visual stimuli varying in complexity -black and white line drawings, coloured line drawings, and arrays of drawings-in participants producing single nouns. Whilst naming latencies were similar for single noun production between visual stimuli conditions, ERPs differed between drawing arrays and single drawings in a time-window extending beyond early visual analysis. In a second experiment, different participants were asked to produce either single noun or adjective-noun dual-word phrases to black-and-white and coloured line drawings, respectively. Adjective-noun phrase production (2W) resulted in naming latencies 53 msec longer than single noun (1W) production. Waveform amplitude and topographic analyses carried out on stimulus- and response-aligned ERPs indicated that the two conditions differed in a late time-window, with a topographic pattern for 2W lasting from 300 to 480 msec after picture presentation whereas the corresponding pattern for 1W production lasted from 300 to 450 msec. Since this time window has been previously associated with phonological encoding in single word production, this result suggests that the cost of planning the second word in dual-word production may be incurred during phonological encoding of the first word. The results are discussed in light of current models of single and multi-word production.