We simulated tsunami propagation for several scenario slip distributions for the 1938 M_W 8.3 earthquake along the Alaska Peninsula, and compared these to the observed records at Unalaska/Dutch Harbor and Sitka. The Sitka record is sensitive to the depth of slip but not the along-strike location, and is fit best by slip at shallow depth. The Unalaska record is sensitive mainly to the along-strike location of slip, and is fit best by slip that is concentrated in the eastern part of the presumed 1938 rupture zone. The tsunami data show that the actual 1938 earthquake rupture zone was smaller than previously thought, likely ~200 km in length, and had no slip near the Shumagin Islands or in the 2020 Simeonof earthquake’s rupture zone. The rupture models that best predict the 1938 tsunami lie within the region of high present day slip deficit inferred from GPS.