Nodal T-follicular helper cell lymphoma (nTFHL) exhibits heterogeneous clinical outcomes that are not fully explained by current clinical and biological variables. While genetic alterations have been extensively investigated, its proteomic landscape remains largely unexplored. Here, we profiled the proteome of 94 nTFHL patients at diagnosis using mass spectrometry coupled with unsupervised machine learning analysis. We identified two nTFHL subtypes with distinct pathway enrichment and significantly different overall survival (p=0.02): the nTHFL-stromal subtype, characterized by a complex microenvironment, and the nTHFL-translational subtype, enriched in protein biosynthesis and cell proliferation pathways. After confirming subtype discrimination across multiple supervised machine learning algorithms, a panel of 10 proteins was selected as biomarkers and validated by immunofluorescence analysis. Among these models, ExtraTrees achieved the highest classification accuracy and was subsequently applied to an independent validation cohort, successfully predicting nTFHL subtypes and confirming the previously observed significant difference in overall survival. In a Cox regression model, the nTHFL-translational subtype remained significantly associated with poorer overall survival compared with the nTFHL-stromal subtype (HR=1.71, p=0.001), independently of the IPI score. Finally, alternative proteins were found to be overexpressed and capable of distinguishing nTFHL subtypes. Overall, our study reveals the heterogeneity of the nTHFL proteome at the time of diagnosis and identifies distinct protein signatures for each subtype, which could be applied in routine practice to support prognostic assessment of nTFHL patients.