Large numbers of hatchery-raised juvenile Pacific salmon are routinely marked with coded-wire tags before release from the hatchery. Log-linear models are an appropriate statistical technique to analyze the numbers of recaptures in different fisheries, and the numbers of tagged fish returning to the hatchery, in terms of such factors as brood year, treatment at the hatchery, timing of release, and size at release. Log-linear analysis of catches and hatchery returns of chinook salmon (Oncorhynchus tshawytscha) from Robertson Creek, British Columbia, indicates that all of these factors are important, but variation between brood years overrides all other factors. Within a brood year, the conditions that maximize the number of returns to the hatchery do not necessarily maximize the number of recaptures in the fishery. Log-linear analysis of hatchery returns from a designed experiment on a single brood year of coho salmon (O. kisutch) from Rosewall Creek, British Columbia, quantifies the effects of the various factors but will be of limited value until the causes of variations between brood years are better known.