This research uses a probabilistic clustering technique to establish a typology of basketball games, considering several performance factors and exogenous variables influencing the probability of winning a game. After analysing NBA data, implementing several filtering processes, and achieving two successive replications, results show that there are five different types of games. The most important variables to discriminate among clusters are the margin of points, the effective field goal percentage and the winning percentage difference between teams. Considerations regarding the home/field advantage, the performance factors, the method to analyse data, and the endogeneity of variables are discussed.