Reports of collapses and whumpfing sounds and their links to instability in the snowpack have been around for decades, but like many areas of snow science, they have received limited scientific treatment. There is a consensus that shear fracture plays an important role in the avalanche process, but the role of collapse is disputed. Many field and laboratory measurements show support for collapse driving fracture, but some refute the idea. The Extended Column Test and the Propagation Saw Test have raised awareness about the role of collapse in crack propagation by showing a lack of sensitivity, in terms of scores, to changes in slope angle – a prediction of the anticrack model. Particle tracking studies focusing on stability tests have likewise shown a strong bending phase that precedes collapse, with no evidence of a sharp shear crack. Yet, crown faces show markings of a fracture that originates at the bed and travels upward, confirming a much older hypothesis that the bed has failed in pure shear, with little or no bending. To understand these contradicting observations, a new theory based on numerical simulations of fracture propagation has recently been proposed and reconciles shear and collapse-based approaches. In this review, we cover theoretical and experimental evidence about the role of collapse in the avalanche failure process as well as theoretical approaches, with an emphasis on research conducted in the past 15 years. We identify gaps and suggest future research.