In the last decade, numerous papers have been delivered on the potential of portable X-Ray Fluorescence (XRF) in archaeological ceramics. Additionally, new chemometric methods have been proposed to manage chemical dataset and facilitate the use of geochemical discrimination for provenance classification of ancient ceramics. In this contribute, the potential of portable Energy Dispersive X-Ray Fluorescence (ED-XRF) analysis and chemical data processing by Graph Clustering is evaluated for provenance classification of archaeological potteries, discussing possible merits and limits of the employed routine. A ceramic assemblage represented by seventy-three transport amphorae classified by typological analysis have been used as testing materials; spectra have been collected on samples simulating in situ analysis conditions (e.g. on fresh cut surfaces without any preparation) and Graph Clustering method has been applied in chemical data processing; comparison with classical Cluster Analysis (CA) and Principal Component Analysis (PCA) is also evaluated. The obtained results favor the use of Graph Clustering for a preliminary classification of ceramics, which can be chemically analyzed in easy, fast and non-destructive way. With a 75.35% of correct attribution, the study shows the suitability of portable ED-XRF in rapid screening of a large number of ceramic samples usually recovered in the framework of archaeological excavation. Misclassifications have been mostly verified for samples exhibiting a coarse-grained clay paste, suggesting that the method is particularly suitable for fine-grained ceramic materials.