Immune checkpoint inhibitors (ICIs) have revolutionized the management of advanced non-small cell lung cancer (NSCLC), although patient survival is still unsatisfactory. Accurate predictive markers capable of personalizing the treatment of patients with NSCLC are still lacking. Circulating extracellular vesicles involved in cell-to-cell communications through miRNAs (EV-miRs) transfer are promising markers. Plasma from 245 patients with advanced NSCLC who received nivolumab as second-line therapy was collected and analyzed. EV-miRnome was profiled on 174/245 patients by microarray platform, and selected EV-miRs were validated by qPCR. A prognostic model combining EV-miR and clinical variables was built using stepwise Cox regression analysis and tested on an independent patient cohort (71/245). EV-PD-L1 gene copy number was assessed by digital PCR. For 54 patients with disease control, EV-miR changes at best response versus baseline were investigated by microarray and validated by qPCR. EV-miRNome profiling at baseline identified two EV-miRs (miR-181a-5p and miR-574-5p) that, combined with performance status, are capable of discriminating patients unlikely from those that are likely to benefit from immunotherapy (median overall survival of 4 months or higher than 9 months, respectively). EV-PD-L1 digital evaluation reported higher baseline copy number in patients at increased risk of mortality, without improving the prognostic score. Best response EV-miRNome profiling selected six deregulated EV-miRs (miR19a-3p, miR-20a-5p, miR-142-3p, miR-1260a, miR-1260b, and miR-5100) in responding patients. Their longitudinal monitoring highlighted a significant downmodulation already in the first treatment cycles, which lasted more than 6 months. Our results demonstrate that EV-miRs are promising prognostic markers for NSCLC patients treated with nivolumab.