Spontaneous activity of neuron populations is often characterized by unique network events, ranging from sharp-wave ripples to slow propagating waves. One of the prominent examples of such network events is population bursting (Fig. 1a), robustly emerging in networks of primary neurons in vitro [1]. Population bursting is believed to depend on the genetic, single neuron, and network features. Typically, the events are characterized via several summary statistics and compared between observations [2]. Although changes in the burst morphology are frequently observed, they are much harder to quantify systematically. Here, using several automatic unsupervised clustering techniques, we show that the composition of burst shapes can be characteristic of a single culture’s genetic background and experimental condition. We extracted individual bursts from a large dataset [3] of developing network cultures and used a hierarchical clustering algorithm (Fig. 1b) to separate them into clusters. To specifically focus on burst shapes, we normalize the duration of each burst and compute the Wasserstein distance between them. Using the Davies-Bouldin index (Fig. 1c), we find that the dataset can be optimally separated into three clusters (Fig. 1c,d). We then study how the bursts change throughout development (Fig. 1e) and we find that the diversity of bursts increases as the cultures develop (Fig. 1f). Thus, the culture’s activity becomes more complex, due to more structured bursting events happening in larger fractions as the culture develops. Moreover, we observe that cultures that share the same genetic background, exhibit similar bursting behavior during the same stage of development. Overall, we find that bursting events become more diverse and complex throughout development. Our results suggest that unsupervised methods can identify structure in the dynamics of neuronal cultures and reveal developmental patterns in one of the most commonly used systems of living neuronal networks.