Abstract: Volatile organic compounds (VOCs) have been widely studied as potential biomarkers of cancer cells. Currently, in vitro cells are mostly cultured in two-dimensional (2D) adherent mode, which is different from in vivo tumor cells with three-dimensional structures. Tumor cell spheres cultured in three-dimensional mode have many similar characteristics to those of in vivo tumors, such as: 1) the structure of cell spheres is divided into proliferative, quiescent, and hypoxic zones; 2) the extracellular matrix deposited in the cell spheres not only increases the density of the spheres but also affects the tumor progression by intercellular interactions. In addition, VOCs detection is mainly accomplished by gas chromatography-mass spectrometry (GC-MS) combined with solid-phase microextraction (SPME), which permits both qualitative and quantitative analysis of VOCs. In this present work, lung cancer cells A549 and lung epithelial cells BEAS-2B were respectively taken as examples to construct three-dimensional (3D) culture models, and 2D models and culture media were used as controls. The VOCs released from cells were detected by SPME-GC-MS. Totally, 163 VOC chromatographic peaks were included in the statistical analyses using an untargeted strategy. Based on orthogonal partial least squares discriminant analysis (OPLS-DA) with the combined criteria of Mann-Whitney U test (p<0.05) and fold change (FC>2), differential VOCs released by cells in 2D and 3D culture modes were screened out. There were four VOCs of acetic acid, 1-pyrroline, 4-methylheptane and 2,4-dimethyl-1-heptene released at differential levels from A549 cells cultured under 2D and 3D modes, and four VOCs of ethanol, 1-pyrroline, 4-methylheptane and 2,4-dimethyl-1-heptene released at various amounts from BEAS-2B cells cultured under 2D and 3D modes, respectively. In comparison to 2D models, 3D models released 2.11 to 12.81 times more of these VOCs. Afterwards, the possible biochemical sources of the differentially released VOCs and the possible mechanisms for the differences in the releases of VOCs in the 2D/3D models were discussed. These differentially released VOCs have been reported to be potential markers in lung cancer cells, liver cancer cells, and other biological samples in previous studies. In addition, compared to 2D models, 3D models can better simulate tumor growth in vivo as well as cellular microenvironments, including the creation of reactive oxygen species (ROS) and extracellular matrix deposition. These physiological microenvironments may promote the release of cellularly differential VOCs. In this paper, by constructing 3D cell models and analyzing differential changes in cellular VOCs, it was proved that 3D culture has a broad application prospect in the detection of cellular VOCs in vitro, and is expected to be a valuable research platform for cancer biomarker screening.