Biological invasions driven by international trade heighten the urgency for development of invasion risk models, as the traits and parameters that consistently predict successful invasion remain unresolved. For four regions of North America that include parts of the United States and Canada (Sacramento-San Joaquin River Basins, Lower Colorado River Basin, Great Lakes Region, Mid-Atlantic Region), we construct and compare classification tree models to reveal robust predictors for the establishment and ecological impact stages of freshwater fish invasion. We subsequently apply the models to identify invasive fish species in trade and conduct pathway analyses to determine which trades (aquarium, biological supply, live bait, live food, water garden) and source continents pose the greatest risk to each region. Model results differed by invasion stage and region. Across regions, establishment models shared climate-related predictors including climate match and temperature tolerance. Three of the four impact models contained prior establishment success. The greatest number of species (548) were predicted to establish in the Sacramento-San Joaquin while the fewest (5) were predicted to establish in the Mid-Atlantic. Forty species were predicted to establish in multiple regions, five of which were also predicted to have high impact. The aquarium trade and Asia supplied the most species predicted to establish. Taken together, the results highlight region-specific models, indicating no universal model predicts invasion. Climate-related and prior establishment variables were most useful to risk assessments. The regional models, and identified high-risk pathways and potential invaders, could be applied to prevent future fish invasions in North America.