This study investigated how much practice is necessary for learners to attain durable second language (L2) grammar knowledge. Using digital flashcards, 119 participants practiced translating 12 sentences into an artificial language, followed by feedback, until they had typed all sentences correctly. Participants repeated this activity in one, two, three, or four relearning sessions on consecutive days. After a 14-day delay, all groups scored highly on a receptive test. However, scores on a productive test were substantially higher for groups with three or four relearning sessions. Accuracy tended to peak on the 3rd day of training. An analysis by individual training performance revealed that participants attained durable productive knowledge if they completed two sessions without errors, regardless of how many sessions they had performed in total. The findings provide a timeframe for processes described in skill retention theory (Kim et al., 2013) and suggest a performance benchmark to indicate when learners have gained procedural L2 grammar knowledge.