Long-term streamflow records are essential for water resources management in semi-arid regions. However, the shortness of instrumental records in Iran and the limited availability of the regional paleoclimate proxies to extend these records back in time hinder our understanding of hydrologic variability. Here, we used the annually resolved tree-ring chronologies from neighboring countries to present the first multi-centennial (~500 years) streamflow reconstruction for the Dez catchment, southwestern Iran, to help understand natural streamflow variability. The Hierarchical Bayesian Regression (HBR) model accompanied by the parsimonious predictor selection method was used to derive the most robust model output. The selection method retained 27 precipitation-sensitive tree-ring chronologies, mainly from regions with similar hydroclimate features to the Dez River catchment. The resulting reconstruction of the lower gauge station accounts for 62% of the variance in streamflow observations, closely matching known historical events and other regional reconstructions. Of particular interest is the recurrence pattern of extreme flow events. Floods were more frequent than low flow extremes (i.e. resembling droughts) during much of the 18th and early 19th centuries. In contrast, droughts were more frequent during the early 17th century.  Worryingly, both flood and drought hazards have increased simultaneously from the beginning of the twentieth century.