Chinook salmon (Oncorhynchus tschawytscha) populations within the highly modified San Francisco Estuary, California, have seen precipitous declines in recent years. To better understand this decline, a decade of coded-wire tag release and recovery data for juvenile salmon was combined with physicochemical data to construct models that represented alternative hypotheses of estuarine conditions that influence tag recovery rate in the ocean. An information theoretic approach was used to evaluate the weight of evidence for each hypothesis and model averaging was performed to determine the level of support for variables that represented individual hypotheses. A single best model was identified for salmon released into the Sacramento River side of the estuary, whereas two competitive models were selected for salmon released into the San Joaquin River side of the estuary. Model averaging found that recovery rates were greatest for San Joaquin River releases when estuary water temperatures were lower, and salmon were released at larger sizes. Recovery rate of Sacramento releases was greatest during years with better water quality. There was little evidence that large-scale water exports or inflows influenced recovery rates in the ocean during this time period. These results suggest that conceptual models of salmon ecology in estuaries should be quantitatively evaluated prior to implementation of recovery actions to maximise the effectiveness of management and facilitate the recovery of depressed Chinook populations.