The current study used a complex, sport-specific movement skill to explore the effects of a win-shift/lose-stay practice schedule on learning and compared its effectiveness with that of blocked and random practice schedules. Thirty-six adults (24.9 ± 3.3 years) were assigned to blocked, random and learner-adapted training groups. Each participant performed 360 trials of the basketball set shot from multiple locations across six practice sessions. For the learner-adapted group, switching between task variations was performance-contingent; switching between shooting locations occurred only following a successful shot. Shooting success was calculated as the percentage of successful shots performed, and measured during pre-acquisition (i.e. baseline), acquisition (i.e. practice) and post-acquisition (i.e. retention and transfer). Despite scoring less shots throughout practice compared to baseline testing, the learner-adapted group showed a significant improvement for shooting success in transfer (d = 1.02). Blocked practice demonstrated significant improvements for shooting success in immediate retention (d = 1.83), delayed retention (d = 1.69) and transfer (d = 1.39). Random practice significantly improved shooting success in both immediate (d = 1.03) and delayed retention (d = 0.81). The current findings highlight that performance during practice does not necessarily represent the permanency and adaptability of skill learning. The implications of the findings and their practical applications are discussed in the context of practice scheduling during learning of a sports skill.