In TBLT, recent years have witnessed sporadic empirical studies probing into the role of working memory in L2 task performance and pedagogy, though a comprehensive picture of their relationship is still not clear (Ellis, Skehan, Li, Shintani & Lambert, 2019). To fill in this gap, the current paper sets out to provide an overview of the differential effects of the two key working memory components (as conceptualized in the Phonological/Executive model, Wen, 2016), either independently or in combination with various L2 task characteristics and task implementation conditions, on various dimensions of L2 speech performance (as indexed by the CAFL framework, Skehan, 2009). As such, I will first recapitulate the three current theoretical models of TBLT from a working memory perspective, namely Peter Skehan’s Limited Attention Capacity Hypothesis (LACH, 2015), Peter Robinson’s Cognition Hypothesis (CH, 2015), and Mike Long’s Interaction Hypothesis (IH, 2014). Then in the second section, I will summarize key results and findings of empirical studies exploring working memory effects on L2 task performance from the four prevailing paradigms of pre-task or strategic planning, online planning, task repetition, as well as studies exploring the role of working memory in the noticing of corrective feedback or recasts. Towards the end, I will draw out the theoretical and pedagogical implications of such a working memory perspective for instructional design in TBLT. (220 words)