In introductory accounting textbooks, virtually all end-of-chapter problems on transaction analysis follow the same familiar format: a collection of transactions performed by a given business during a specified time period. Modern research-based models of human cognitive architecture suggest, however, that this format is suboptimal for beginning students. An approach better aligned with this learning research would give students practice with one transaction type at a time before proceeding to problems involving a mixture of transaction types. An experiment was conducted to test this hypothesis by randomly assigning students in an introductory financial accounting course to one of two practice conditions: conventional textbook problems and “targeted practice” in which the same transactions were grouped by type. All students were then given a conventional textbook problem as a post-test. During the practice phase, students in the targeted practice group analyzed transactions in less time and with greater accuracy than students who worked conventional problems. On the post-test, the total scores of the two groups were statistically equivalent; thus, the targeted practice group achieved the same level of performance more efficiently. However, on transactions requiring transfer of learning, the targeted practice group performed notably better, indicating these students were better able to apply knowledge gained during practice to a broad variety of transaction scenarios. The implications of this study are straightforward and practical: by making a very simple modification to the format of transaction analysis problems given to students early in the learning process, better learning outcomes can be obtained.