The tree-ring signal for flooding along the Ob River, a large Arctic River in western Siberia, is investigated using a combination of floodplain tree-ring sites from riparian and non-riparian settings. A conceptual model is presented contrasting tree-growth responses of riparian and non-riparian trees to unusually severe flooding. A set of five riparian (Salix and Populus) tree-ring chronologies is developed and used in combination with existing floodplain non-riparian Larix and Pinus chronologies in a binary classification tree (CT) model to classify high-flood years, defined as a Salekhard water-level gage reading in the seasonal window from May 1 to August 31 of above 470 cm for 82 or more consecutive days. Correlation and regression identifies a nonlinear relationship of riparian ring widths to discharge and flooding: higher annual discharge generally leads to higher growth, but the relationship reverses in extreme-flood years. Micrographs highlight the suppression of width and occasional distortion of cell anatomy in selected trees. CT modeling guided by cross-validation yields a CT model with a primary split on the riparian ring width and secondary split on the non-riparian ring width. The model successfully identifies four of the eight most severe high-flow years, 1934–2014. The model further identifies two years (1885 and 1914) before the start of the gaged record in 1934 as high-flow years. No appreciable difference is found in frequency of high-flow years before and after 1956, when the first major reservoirs began filling upstream of the Lower Ob. The CT modeling approach is proposed as a novel approach to dealing with nonlinearity in reconstructing flood history of Arctic rivers from tree rings.