OBJECTIVES: To identify plasma extracellular vesicles (EVs) associated with radiographic knee osteoarthritis (OA) progression. METHODS: EVs of small (SEV), medium (MEV) and large (LEV) sizes from plasma of OA participants (n=30) and healthy controls (HCs, n=22) were profiled for surface markers and cytokine cargo by high-resolution flow cytometry. The concentrations of cytokines within (endo-) and outside (exo-) EVs were quantified by multiplex ELISA. EV associations with knee radiographic OA (rOA) progression were assessed by multivariable linear regression (adjusted for baseline clinical variables of age, gender, BMI and OA severity) and receiver operating characteristic (ROC) curve analysis. RESULTS: Based on integrated mean fluorescence intensity (iMFI), baseline plasma MEVs carrying CD56 (corresponding to natural killer cells) predicted rOA progression with highest area under the ROC curve (AUC) 0.714 among surface markers. Baseline iMFI of TNF-α in LEVs, MEVs and SEVs, and the total endo-EV TNF-α concentration, predicted rOA progression with AUCs 0.688, 0.821, 0.821, 0.665, respectively. In contrast, baseline plasma exo-EV TNF-α (the concentration in the same unit of plasma after EV depletion) did not predict rOA progression (AUC 0.478). Baseline endo-EV IFN-γ and exo-EV IL-6 concentrations were also associated with rOA progression, but had low discriminant capacity (AUCs 0.558 and 0.518, respectively). CONCLUSIONS: Plasma EVs carry pro-inflammatory cargo that predict risk of knee rOA progression. These findings suggest that EV-associated TNF-α may be pathogenic in OA. The sequestration of pathogenic TNF-α within EVs may provide an explanation for the lack of success of systemic TNF-α inhibitors in OA trials to date.