We analyze a large database of abyssal peridotite clinopyroxene compositions using log-ratio transformation, principal component analysis (PCA) and k-means clustering, to better understand clinopyroxene compositional systematics in abyssal peridotites. We combine this analysis with open-system melting models to investigate the potential sources of compositional variation. PCA shows that 84% of the variation in clinopyroxene compositions can be represented using only 2-dimensional information. We use k-means clustering to classify clinopyroxene compositions into four clusters. Clusters 1–3, representing 85% of the data, show progressive depletions in LREE/HREE, and are associated with decreases in Na2O in clinopyroxene, and general increases in Cr# of spinel. We interpret peridotites with clinopyroxene compositions from clusters 1–3 to represent residues of partial melt extraction. The degree of melt extraction increases from cluster 1 to 3, and exerts a primary control on compositional variations. The presence or absence of garnet-field melting prior to spinel-field melting and the retained melt fraction during partial melting exert secondary controls on clinopyroxene compositions. Cluster 4 clinopyroxenes, representing show less fractionated LREE/HREE with low-HREE abundances, elevated Sr, and depleted signatures in their host peridotites. Clinopyroxene compositions in cluster 4 cannot be modeled by melt depletion alone. Instead, they are only reconstructed in our models where melt-rock interaction, suggesting that peridotites with cluster 4 clinopyroxenes have experienced both of these processes. Clusters 1–4 are observed in most ridges, reflecting compositional heterogeneity on each ridge. This variability reflects variations in the degree of partial melting, amount of garnet-field melting, retained melt fractions, and melt-rock interaction.