Epithelial-mesenchymal transition (EMT) and cancer stem-like cells (CSC) are becoming highly relevant targets in anticancer drug discovery. A large body of evidence suggests that epithelial-mesenchymal transitioned tumor cells (EMT tumor cells) and CSCs have similar functions. There is also an overlap regarding the stimuli that can induce the generation of EMT tumor cells and CSCs. Moreover, direct evidence has been brought that EMT can give rise to CSCs. It is unclear however, whether EMT tumor cells should be considered CSCs or if they have to undergo further changes. In this article we summarize available evidence suggesting that, indeed, additional programs must be engaged and we propose that macroautophagy (hereafter, autophagy) represents a key trait distinguishing CSCs from EMT tumor cells. Thus, CSCs have often been reported to be in an autophagic state and blockade of autophagy inhibits CSCs. On the other hand, there is ample evidence showing that EMT and autophagy are distinct events. CSCs, however, represent, by themselves, a heterogeneous population. Thus, CSCs have been distinguished in predominantly non-cycling and cycling CSCs, the latter representing CSCs that self-renew and replenish the pool of differentiated tumor cells. We now suggest that the non-cycling CSC subpopulation is in an autophagic state. We propose also two models to explain the relationship between EMT tumor cells and these two major CSC subpopulations: a branching model in which EMT tumor cells can give rise to cycling or non-cycling CSCs, respectively, and a hierarchical model in which EMT tumor cells are first induced to become autophagic CSCs and, subsequently, cycling CSCs. Finally, we address the therapeutic consequences of these insights.