Three-dimensional (3D) cancer tumor models are becoming vital approaches for high-throughput drug screening, drug targeting, development of novel theranostic systems, and personalized medicine. Yet, it is becoming more evident that the tumor progression and metastasis is fueled by a subpopulation of stem-like cells within the tumor that are also called cancer stem cells (CSCs). This study aimed to develop a tumoroid model using CSCs. For this purpose CD133(+) cells were isolated from SaOS-2 osteosarcoma cell line with magnetic-activated cell sorting. To evaluate tumoroid formation ability, the cells were incubated in different cell numbers in agar gels produced by 3D Petri Dish (R) method. Subsequently, CD133(+) cells and CD133(-) cells were co-cultured to investigate CD133(+) cell localization in tumoroids. The characterization of tumoroids was performed using Live&Dead staining, immunohistochemistry, and quantitative polymerase chain reaction analysis. The results showed that, CD133(+), CD133(-) and SaOS-2 cells were all able to form 3D tumoroids regardless of the initial cell number, but, while 72 hr were needed for CD133(+) cells to self-assemble, 24 hr were enough for CD133(-) and SaOS-2 cells. CD133(+) cells were located within tumoroids randomly with high cell viability. Finally, when compared to two-dimensional (2D) cultures, there were 5.88, 4.14, 6.95, and 1.68-fold higher messenger RNA expressions for Sox2, OCT3/4, Nanog, and Nestin, respectively, in CD133(+) cells that were cultured within 3D tumoroids, showing longer maintenance of stem cell phenotype in 3D, that can allow more relevant screening and targeting efficiency in pharmaceutical testing. It was concluded that CSC-based tumoroids are propitious as 3D tumor models to fill the gap between conventional 2D in vitro culture and in vivo animal experiments for cancer research.