Cells generate forces in tissues to sculpt the myriad of tissue shapes that occur in an organism. An engine that is associated with cell and tissue shape changes in diverse contexts is a dynamic actomyosin meshwork or cortex underlying the plasma membrane. In epithelial cells, contraction of the apical actomyosin cortex can drive apical constriction. How apical actomyosin meshworks contract is not well understood. In particular, the organization and polarity of actin filaments in the cortex are not well defined in contractile epithelial cells. Here, we show that in constricting non-muscle cells of the Drosophila embryo, actin filament pointed ends and myosin are enriched near the center of the apical cortex and actin filament barbed ends are enriched at the junctions. Thus, the apical cortex of non-muscle cells can adopt a polarized structure that topologically resembles a muscle sarcomere. Importantly, we provide evidence that depolarized apical Rho-kinase (ROCK) activity is insufficient to contract actin networks or to promote apical constriction, suggesting that targeted myosin activation by ROCK at the center of the cell apex is required for cell shape change. Polarized ROCK accumulation requires dynamic regulation of its upstream activator RhoA, and we identified an uncharacterized RhoA GAP that is critical for establishing ROCK polarity and apical constriction. ROCK activity is continuously required to sustain its own polarity and also the junctional enrichment of E-cadherin within the apical cortex, demonstrating ROCK’s key role in organizing a contractile apical cortex. In conclusion, we identified that establishing ROCK polarity and a sarcomere-like organization to the apical cortex requires a dynamic upstream signaling network that surprisingly involves, not only activation, but inhibition of the RhoA GTPase.