Non-invasive assays for early cancer screening are hampered by challenges in the isolation and profiling of circulating biomarkers. Here, we show that surface proteins from serum extracellular vesicles labelled with a panel of seven fluorescent aptamers can be profiled, via thermophoretic enrichment and linear discriminant analysis, for cancer detection and classification. In a cohort of 102 patients, including 6 cancer types at stages I-IV, the assay detected stage I cancers with 95% sensitivity (95% confidence interval (CI): 74-100%) and 100% specificity (95% CI: 80-100%), and classified the cancer type with an overall accuracy of 68% (95% CI: 59-77%). For patients who underwent prostate biopsies, the assay was superior to the analysis of prostate-specific antigen levels (area under the curve: 0.94 versus 0.68; 33 patients) for the discrimination of prostate cancer and benign prostate enlargement, and also in the assessment of biochemical cancer recurrence after radical prostatectomy. The assay is inexpensive, fast, and requires small serum volumes (<1 µl), and if validated in larger cohorts may facilitate cancer screening, classification and monitoring.