Background Reverse total shoulder arthroplasty (RTSA) is a treatment option for patients with severe osteoarthritis, rotator cuff arthropathy, or massive rotator cuff tear with pseudoparalysis. We are to deduce not only the early functional outcomes and complications of cementless RTSA during the learning curve period but also complication-based, and operation time-based learning curve of RTSA. Methods Between March 2010 and February 2014, we retrospectively evaluated 38 shoulders (6 male, 32 female). The average age of the patients was 73.0 years (range, 63 to 83 years), and the average follow-up was at 24 months (range, 12–53 months). The visual analog scale (VAS), University of California Los Angeles (UCLA) score and constant score were used to evaluate the clinical outcomes. We evaluated patients radiographically at 2 weeks, 3 months, 6 months, 1 year, and then annually thereafter for any evidence of complications. Results The VAS score improved from 4.0 to 2.8 ( p = 0.013). The UCLA score improved from 16.0 to 27.9 ( p = 0.002), and the constant score improved from 41.4 to 78.9 ( p < 0.001), which were statistically significant. While active forward flexion, abduction, and internal rotation improved ( p value = 0.001, < 0.01, 0.15), external rotation did not show significant improvement ( p = 0.764). Postoperative complications included acromion fracture (one case), glenoid fracture (one case), peripristhetic humeral fracture (one case), axillary nerve injury (one case), infection (one case), and arterial injury (one case). Our study presented an intraoperative complication-based learning curve of 20 shoulders, and operation time-based learning curve of 15 shoulders. Conclusions The clinical outcomes of RTSA were satisfactory with overall complication rates of 15.7%. An orthopedic surgeon within the learning curve period for the operation of RTSA should be cautious when selecting the patients and performing RTSA. Trial registration Retrospectively registered.