Mastering a cognitive skill requires many practice sessions, occurring over a period of days, weeks, months, or even years. Although a large body of research describes and explains gains made within a given practice session, few studies have investigated what happens to these gains across a delay, and none have examined effects of delays on item-general gains. Across 3 experiments, participants performed alphabet arithmetic verification in an initial practice session followed by a test session after a delay (from 0 to 30 days). All experiments included conditions yielding item-general practice gains; Experiments 2-3 also included an item-specific practice condition. Surprisingly, item-general gains were relatively well preserved across a delay (e.g., only 6.7% decrease in practice effects after 2 days), whereas item-specific gains showed sizeable losses across a delay (e.g., 25.9% loss after 2 days). Results provide important empirical constraints to theories of cognitive skill acquisition.