This study aimed to explore which offensive variables best discriminate the ball possession duration according to players specific role (defenders, midfielders, attackers) during a Gk+3vs3+Gk football small-sided games. Fifteen under-15 players (age 13.2±1.0 years, playing experience 4.2±1.0 years) were grouped according to their positions (team of defenders, n = 5; team of midfielders, n = 7; team of attackers, n = 3). On each testing day (n = 3), each team performed one bout of 5-min against each team in a random order, accounting for a total of nine bouts in the following scenarios: i) defenders vs midfielders; ii) defenders vs attackers; iii) midfielders vs attackers. Based on video, a notational analysis process allowed to capture individual and collective actions. According to each playing position group, discriminant analysis was used to identify relevant variables that discriminate different ball possession sequences (short, medium, and long). The analysis revealed the existence of three clusters according to ball possession duration, classified as short sequence (~4 seconds), medium sequence (~10 seconds) and long sequence (~18 seconds). The number of touches per possession was the variable that discriminates the ball possession duration from all playing positions while passing actions were related to midfielders and attackers. In addition, different ball possessions sequences in the attackers were also discriminated by the number of players involved per possession. Accordingly, to increase the duration of the offensive phase during small-sided games, coaches should foster the players' ability to stay on the ball, as it may amplify their opportunities to maintain the ball possession. In addition, coaches may also include reward rules to encourage midfielders and attackers' passing actions and the number of attackers involved during the attack to promote longer ball possessions durations.