Previous experiments have shown a deleterious effect of visual representations on college students' ability to solve total- and joint-probability word problems. The present experiments used conditional-probability problems, known to be more difficult than total- and joint-probability problems. The diagram group was instructed in how to use tree diagrams to solve conditional probability problems; the equation-only group was instructed only in the formulation of equations. Results indicated that the diagram group outperformed the equation-only group on a posttest of 10 near- and far-transfer problems given immediately and as long as 1 week later (Experiment 2). In addition, when rating their experience with the training phase the equation-only group reported significantly higher levels of frustration and lower levels of perceived success than the diagram group.