Introduction: Researchers have widely investigated how interventions by means of training can improve manual behaviors in infants. However, no systematic review has been found on this topic. Objective: To analyze the quality of scientific evidence considering the methodological quality and level of evidence by type of study in research on training of object-directed manual behaviors in infants in the first 18 months of life. Methods: National Library of Medicine (PubMed/MEDLINE), Latin American and Caribbean Health Sciences (LILACS), Virtual Health Library (BIREME/BVS), Science Direct, SciELO, and Physiotherapy Evidence Database (PEDro) databases were used. Only clinical trials that assessed the benefits of manual object-directed training in infants and were published up to February 2018, in English, were included. The Cochrane Collaboration Model was adapted to extract bibliographical data from the articles and their methodological quality was assessed using the PEDro scale and the Oxford Centre for Evidence-Based Medicine’s Levels of Evidence. Results: Twenty one clinical trials were included. Studies investigated typically developing full-term infants, preterm infants, and infants at risk for autism spectrum disorders. Trainings were administered to infants by means of “sticky mittens” paradigm, task-specific practice, or contingency reinforcement. Most of the studies presented fair or poor methodological quality. Only studies that used task-specific active practice presented high methodological quality. Conclusions: The results indicate there is high quality evidence that task-specific training improves object-directed manual behaviors in typically developing infants and preterm infants in the first 2-4 months of life. Studies addressing infants with established diagnoses of developmental dysfunction are lacking.