This study proposes a novel approach to clustering football teams’ playing styles using a Deep Embedded Clustering (DEC) algorithm applied to large-scale event data. The dataset comprises over 3 million events from 1,826 matches played in the top-tier leagues of England, Spain, Italy, Germany, and France during the 2016/2017 season. Each match event, such as passes, shots, and duels, contributes to a comprehensive representation of team behavior on the field. To capture tactical nuances, the dataset is segmented into four distinct phases of play, and each phase is clustered independently. These intermediate clustering results are aggregated to create a feature representation for each team, which is subsequently clustered to reveal dominant playing styles. A detailed feature engineering process, inspired by recent literature, incorporates spatial and temporal elements of play, including pass motifs, positional tendencies, and graph-based metrics. The resulting clusters are evaluated in terms of their clustering quality, measured by Silhouette, \(A(C)_1\) , and \(A(C)_2\) scores, their predictive utility for match outcomes, and their tactical interpretability. The analysis demonstrates clear performance disparities among styles, offering insights into the effectiveness of specific tactical schemas. This methodology enables data-driven tactical analysis and benchmarking in football, highlighting the potential of unsupervised learning, particularly deep clustering, to inform strategic decision-making in sports analytics. The source code of this study is available at: https://github.com/egecjdemir/how_football_teams_play .