ABSTRACT In late Wisconsin time, the Purcell Trench lobe of the Cordilleran ice sheet dammed the Clark Fork of the Columbia River in western Montana, creating glacial Lake Missoula. During part of this epoch, the Okanogan lobe also dammed the Columbia River downstream, creating glacial Lake Columbia in northeast Washington. Repeated failure of the Purcell Trench ice dam released glacial Lake Missoula, causing dozens of catastrophic floods in eastern Washington that can be distinguished by the geologic record they left behind. These floods removed tens of meters of pale loess from dark basalt substrate, forming scars along flowpaths visible from space. Different positions of the Okanogan lobe are required for modeled Missoula floods to inundate the diverse channels that show field evidence for flooding, as shown by accurate dam-break flood modeling using a roughly 185 m digital terrain model of existing topography (with control points dynamically varied using automatic mesh refinement). The maximum extent of the Okanogan lobe, which blocked inundation of the upper Grand Coulee and the Columbia River valley, is required to flood all channels in the Telford scablands and to produce highest flood stages in Pasco Basin. Alternatively, the Columbia River valley must have been open and the upper Grand Coulee blocked to nearly match evidence for high water on Pangborn bar near Wenatchee, Washington, and to flood Quincy Basin from the west. Finally, if the Columbia River valley and upper Grand Coulee were both open, Quincy Basin would have flooded from the northeast. In all these scenarios, the discrepancy between modeled flood stages and field evidence for maximum flood stages increases in all channels downstream, from Spokane to Umatilla Basin. The pattern of discrepancies indicates that bulking of floods by loess increased flow volume across the scablands, but this alone does not explain low ­modeled flow stages along the Columbia River valley near Wenatchee. This latter discrepancy between modeled flood stages and field data requires either additional bulking of flow by sediment along the Columbia reach downstream of glacial Lake Columbia, or