It remains unclear whether and to what extent working memory (WM) temporal subprocesses (i.e., encoding, maintenance, and retrieval) involve shared or distinct intrinsic networks. To address this issue, I constructed a model of intrinsic network contributions to different WM phases and then evaluated the validity of the model by performing a quantitative meta-analysis of relevant functional neuroimaging data. The model suggests that the transition from the encoding to maintenance and to retrieval stages involves progressively decreasing involvement of the dorsal attention network (DAN), but progressively increasing involvement of the frontoparietal control network (FPCN). Separate meta-analysis of each phase effect and direct comparisons between them yielded results that were largely consistent with the model. This evidence included between-phase double dissociations that were consistent with the model, such as encoding > maintenance contrast showing some DAN, but no FPCN, regions, and maintenance > encoding contrast showing the reverse, that is, some FPCN, but no DAN, regions. Two closely juxtaposed regions that are members of the DAN and FPCN, such as inferior frontal junction versus caudal prefrontal cortex and superior versus inferior intraparietal sulcus, showed a high degree of functional differentiation. Although all regions identified in the present study were already identified in previous WM studies, this study uniquely enhances our understating of their roles by clarifying their network membership and specific associations with different WM phases.