We ask how complex (dual-action) behavior is mentally represented in dual tasks. Three different representation accounts were empirically tested. According to a compositional (Structuralist) account, component action remains structurally intact when combined with another action. In contrast, a holistic (Gestalt) account posits that dual-action requirements are represented holistically and entirely distinct from its component action requirements. Finally, a contextual change account assumes that a change in context (e.g., from single- to dual-action requirement) generally impedes response retrieval, similar to repeating a response while the task context switches. We analyzed trial-by-trial effects in a single/dual switch paradigm (SDS paradigm, involving a randomized mix of single- and dual-task trials) combining an auditory-vocal task and a visual-manual task. Relevant comparisons of performance between complete switch trials (e.g., between the two single tasks) and partial repetition trials (e.g., from dual to single task) revealed partial repetition benefits, that is, for both the auditory-vocal and the visual-manual task, and for both single- and dual-task performance. For example, executing a manual response in a current dual-task trial benefitted from having executed a manual response in the previous single-task trial. The pattern of results persisted across several days of performance. Therefore, dual actions in the present dual-task setting are mentally represented in a compositional, Structuralist fashion throughout, likely resulting from low between-task dimensional overlap. The results will be discussed within a theoretical framework of flexible representation of behavior.