Cancer stem cells (CSCs) play critical roles in cancer, making them important targets for new diagnostic and therapeutic approaches. Since CSCs are heterogeneous and not abundant in tumors, and few specific markers for these cells currently exist, new methods to isolate and characterize them are required. To address this issue, we developed a new label-free methodology to isolate, enrich, and identify CSCs from an heterogeneous tumor cell subpopulation using a cell sorting method (sedimentation field flow fractionation, SdFFF) and a biosensor as a detector. Enrichment was optimized using an original protocol and U87-MG glioblastoma cells cultured in a normal (N) or defined (D) medium (± fetal bovine serum, FBS) under normoxic (N, pO2 = 20%) or hypoxic (H, pO2 < 2%) conditions to obtain four cell populations: NN, NH, DN, and DH. After elution of CSCs via SdFFF using the hyperlayer mode (inertial elution mode for micrometer-sized species), we isolated eight subpopulations with distinct CSC contents based on phenotypical and functional properties, ranging from NN F1 with a lower CSC content to DH F3 with a higher CSC content. Reflecting biological differences, the intrinsic intracellular dielectric permittivity increased from NN to DH conditions. The largest difference in electromagnetic signature was observed between NN F1 and DH F3, in which the CSC content was lowest and highest, respectively. The results demonstrate that microwave dielectric spectroscopy can be used to reliably and efficiently distinguish stem cell characteristics. This new instrumental and methodological approach is an important innovation that allows both enrichment and detection of CSCs, opening the door to novel diagnostic and therapeutic approaches.