Contemporaneous mass-transport deposits (MTDs) recorded in lake and fjord sediments provide evidence of past seismic shaking. However, because they are usually not connected to a fault rupture, assessment of the earthquake source remains difficult. Based on observed coseismic mass wasting and associated seismic shaking, previous studies assigned minimum intensities required to trigger them. Attempts to infer their earthquake source relied on methods developed to estimate the location and magnitude of historical earthquakes using intensity prediction equations, but considered these thresholds as actual intensities. Here we develop a probabilistic method to infer the most likely earthquake sources from the spatial distribution (or absence) of MTDs. This approach simultaneously allows the triggering intensity to exceed the assumed threshold and takes into account intensity prediction equation uncertainties, two shortcomings of existing methods. Additionally, we consider known active faults rather than a grid of possible epicenters. We apply this method to Aysen Fjord (southern Chile), which is intersected by the Liquine-Ofqui Fault Zone. In 2007, an M-W=6.2 earthquake hit the fjord with intensities of VIII+, causing major landslides entering the fjord. Seismic-reflection profiles show that its sedimentary fill contains nine prehistoric MTD levels. Following a sensitivity analysis, application of the method to the MTD record allows identifying the most likely fault sections and magnitude range for most events, confirming that they are mainly attributed to crustal earthquakes on the Liquine-Ofqui Fault Zone. We conclude that the method has good potential to constrain the size and location of paleoearthquakes for which only shaking evidence is available. Plain Language Summary Strong earthquakes often cause landslides that may be preserved as typical deposits in lakes and fjords. Previous studies have linked the triggering of different types of landslides to minimum shaking intensities and used empirical relations with earthquake magnitude and distance in a simple way to derive the most likely location and magnitude of the earthquake. We developed an alternative method that takes into account the uncertainties in these intensity relations and also the possibility that the actual intensity was higher than the minimum triggering intensity. We extend the method by using information on known active faults in the area, and apply it to a fjord in Chile that is intersected by several faults. An earthquake on one of these faults in 2007 caused various landslides, which are preserved in the fjord's sedimentary record, along with nine older landslide events. In most cases, our method appears to perform better than previous methods to identify the faults that most likely caused the observed landslide pattern. Our results confirm that most landslides are the result of earthquakes on faults in or around the fjord and that only a few could be the result of larger earthquakes on the subduction zone off the coast.