Abstract Fireball networks are used to recover meteorites, with the context of orbits. Observations from these networks cover the bright flight, where the meteoroid is luminescent, but to recover a fallen meteorite, these observations must often be predicted forward in time to the ground to estimate an impact position. This dark-flight modeling is deceptively simple, but there is hidden complexity covering the precise interactions between the meteorite and the (usually active) atmosphere. We describe the method and approach used by the Desert Fireball Network, detailing the issues we have addressed, and the impact that factors such as shape, mass, and density have on the predicted fall position. We illustrate this with a case study of Murrili meteorite fall that occurred into Lake Eyre-Kati Thanda in 2015. The fall was very well observed from multiple viewpoints, and the trajectory was steep, with a low-altitude endpoint, such that the dark flight was relatively short. Murrili is 1.68 kg with a typical ordinary chondrite density but with a somewhat flattened shape compared to a sphere, such that there are discrepancies between sphere-based predictions and the actual recovery location. It is notable that even in this relatively idealized dark-flight scenario, modeling using spherically shaped projectiles resulted in a significant distance between predicted fall position and recovered meteorite.