Calderas are volcanic depressions formed when the roof of a magma chamber collapses due to the depletion of magma within the chamber. Investigating the development of caldera boundary faults that accommodate chamber-roof subsidence is crucial for understanding caldera collapse events and predicting contemporaneous and subsequent volcanic eruptions. Because subsurface structures in natural calderas are difficult to observe, analogue sandbox and numerical experiments, and inversions of geodetic data are often used to reconstruct caldera structures. Recent experimental and analytical studies have revealed localized downsag during the upward propagation of caldera boundary faults and that the fault structures of trapdoor calderas are highly variable. However, how downsag becomes localized during upward boundary fault propagation, and when and why trapdoor structures originate during caldera collapse remain open questions. Here, we performed analogue sandbox experiments in a transparent sandbox and used image analysis techniques to provide insights into these questions. Calderas with 1–10 km in diameter are more favorable than larger ones (tens of kilometers in diameter) for assessing downsagging and trapdoor subsidence, because these volcanic deformations may be masked by crustal deformation due to regional tectonics controlling large caldera subsidence. We therefore focused on calderas with 1–10 km in diameter, and excluded some factors such as pre-existing stress regimes, faults and magma flow, in order to evaluate the development of downsag and trapdoor subsidence themselves. Despite these experimental limitations, our experiments are in agreement with natural calderas and show that as caldera subsidence increases, downsag and trapdoor faulting evolve in four stages. (1) At the beginning of downsag, the horizontal displacement velocity is greatest along the edge of the downsagging region. (2) As the outward-dipping reverse faults nucleate at depth and propagate upward, the peak of horizontal displacement velocity localizes along the surface projection of the concealed faults. (3) Caldera collapse blocks then undergo trapdoor subsidence when the outward-dipping reverse faults reach the surface. (4) Finally, a second episode of trapdoor subsidence of the collapse block occurs when inward-dipping normal faults nucleate beyond the outward-dipping reverse faults. The first and second stages suggest that the existence and location of concealed caldera boundary faults can be predicted from the downsag deformation pattern. The third and fourth stages explain temporal and structural variations observed in trapdoor calderas in the context of caldera boundary fault development. Our model will improve the resolution of caldera structural reconstructions and associated inferences regarding magma chamber dynamics.