Understanding and optimizing spacing during learning is a central topic for research in learning and memory and has substantial implications for real-world learning. Spacing memory retrievals across time improves memory relative to massed practice—the well-known spacing effect. Most spacing research has utilized fixed (predetermined) spacing intervals. Some findings indicate advantages of expanding over equal spacing (e.g., Landauer & Bjork, 1978); however, evidence is mixed (e.g., Karpicke & Roediger, 2007), and the field has lacked an integrated explanation. Learning may instead depend on interactions of spacing with an underlying variable of learning strength that varies for learners and items, and it may be better optimized by adaptive adjustments of spacing to learners’ ongoing performance. Two studies investigated an adaptive spacing algorithm, Adaptive Response-Time-based Sequencing or ARTS (Mettler, Massey & Kellman, 2011) that uses response-time and accuracy to generate spacing. Experiment 1 compared adaptive scheduling with fixed schedules having either expanding or equal spacing. Experiment 2 compared adaptive schedules to 2 fixed “yoked” schedules that were copied from adaptive participants, equating average spacing across conditions. In both experiments, adaptive scheduling outperformed fixed conditions at immediate and delayed tests of retention. No evidence was found for differences between expanding and equal spacing. Yoked conditions showed that learning gains were due to adaptation to individual items and learners. Adaptive spacing based on ongoing assessments of learning strength yields greater learning gains than fixed schedules, a finding that helps to understand the spacing effect theoretically and has direct applications for enhancing learning in many domains.