Voluntary movement and heartbeat in animals are powered by the contraction of periodically patterned acto-myosin bundles, called myofibrils, which are present inside striated skeletal and cardiac muscle cells. Myofibrils are “active cytoskeletal crystals” that span across the long muscle cells with hundreds of stereotyped, micrometer-sized sarcomere units repeated in series. Despite enormous progress in understanding the molecular composition and structure of sarcomeres, we still know little about the physical mechanisms that underlie sarcomere assembly during development. In this review, we critically discuss different hypotheses on the biophysical mechanisms of myofibrillogenesis. We describe how periodic myofibrils originate from mixed-polarity acto-myosin bundles that resemble unstriated stress fibers. These bundles undergo a transition from nematic (parallel) to smectic (layered) order, thereby establishing a state of higher molecular order, with alternating localization of myosin and actin crosslinkers. This order transition from a spatially homogeneous to a periodic pattern occurs simultaneously across the long myofibrils and is likely coordinated by mechanical tension. We provide an overview of previous theoretical models of sarcomere assembly. In light of recent experimental advances, we propose a new model in which muscle-specific myosin motor filaments, actin crosslinkers, and titin interact in a feedback loop that establishes an initial periodic pattern of proteins bound to a mixed-polarity actin filament bundle. This initial periodic pattern guides subsequent remodeling and results in a polarity sorting of the actin filaments, remodeling the nematic (mixed polarity) actin filament bundle into a smectic ordered chain of mechanically connected actin filament stacks with alternating polarity. This new model integrates previously contradictory ideas, such as the prominent pre-myofibril hypothesis, the stitching model, and the concept of tension-driven sarcomere self-organization.