Anticancer drug discovery has been hampered by the lack of reliable preclinical models, which routinely use cells grown in two-dimensional (2D) culture systems. However, many of the characteristics of cells in 2D culture do not translate into the findings in animal xenografts. Three-dimensional (3D) growth may be responsible for some of these changes, and models using cells grown in 3D may form a more representative step in tumouricidal validation prior to animal implantation and human testing. For the 3D model, we cultured 143.98.2, SaOS2 or U2OS osteosarcoma cells seeded in porous Bombyx mori silk sponges. We conducted real-time PCR on cells grown in 2D culture and 3D scaffolds for the proliferation markers cyclin B1 and E2F1 and the actin regulator RhoA, and found a significant decrease in expression levels for the 3D tumour models (p = 0.02, < 0.001 and 0.008 for cyclin B1, E2F1 and RhoA for 143.98.2; p = 0.02, 0.002 and 0.02 for cyclin B1, E2F1 and RhoA for U2OS, respectively). In contrast, p21 was upregulated when SaOS2 and U2OS were cultured in the 3D scaffolds (p < 0.001) and there was no increase in DNA quantity during the culture period. We correspondingly observed G1 arrest when cell cycle analysis was conducted. Cytotoxicity results for cells treated with serial dilutions of doxorubicin and cisplatin showed that cells in 3D scaffolds were less sensitive to drug treatment than in 2D culture, and the difference was more pronounced for cell cycle specific agents. Copyright © 2013 John Wiley & Sons, Ltd.