Declines in cottonwood (Populus spp.) recruitment along alluvial reaches of large rivers in arid regions of the western United States have been attributed to modified flow regimes, lack of suitable substrate, insufficient seed rain, and increased interspecific competition. We evaluated whether and how these factors were operating during 1993–1996 to influence demographics of Fremont cottonwood (P. deltoides Marshall subsp. wislizenii (Watson) Eckenwalder) along reaches of the Green and Yampa Rivers near their confluence in northwestern Colorado. We examined seedling establishment, defined as survival through three growing seasons, at three alluvial reaches that differed primarily in the level of flow regulation: a site on the unregulated Yampa, an upper Green River site regulated by Flaming Gorge Dam, and a lower Green River site below the Green–Yampa confluence. Seed rain was abundant in all sites, and led to large numbers of germinants (first-year seedlings) appearing each year at all sites. The regulated flow in the upper Green River reach restricted germination to islands and cut banks that were later inundated or eroded; no seedlings survived there. Mortality at the lower Green River site was due largely to desiccation or substrate erosion; 23% of 1993 germinants survived their first growing season, but at most 2% survived through their second. At the Yampa River site, germinants appeared on vegetated and unvegetated surfaces up to 2.5 m above base flow stage, but survived to autumn only on bare surfaces at least 1.25 m above base flow stage, and where at least 10 of the upper 40 cm of the alluvium was fine-textured. Our studies of rooting depths and the stable isotopic composition of xylem water showed that seedlings in the most favorable locations for establishment at the Yampa site do not become phreatophytic until their third or fourth growing season. Further, the results of experimental field studies examining effects of shade and competition supported the hypothesis that insufficient soil moisture, possibly in combination with insufficient light, restricts establishment to unvegetated sites. Collectively, the demographic and experimental studies suggest that, in arid regions, soil water availability is at least as important as light level in limiting establishment of Fremont cottonwood seedlings. We hypothesize that in cases where arid land rivers experience large spring stage changes, recruitment is further constrained within bare areas to those sites that contain sufficient fine-textured alluvium, saturated during the spring flood, to provide the flood-derived soil moisture normally necessary for late-summer seedling survival.