In this study, the effects of threshold variation in image segmentation of micro CT images of cancellous bone in the determination of the architectural parameters and stiffness were investigated. A total of 42 samples of 6 × 6 × 6 mm3 cubes with threshold values set between 500–1100 greyscale in increment of 100 of CT images of six human C5 vertebral bodies were analyzed. Threshold value of 800, based on Otsu's method, was set for the control group. From various threshold values, the respective architectural parameters, and the corresponding stiffness in three orthotropic directions (Exx, Eyy, Ezz) of each cube were computed from the voxel-based micro-finite element models under compressive simulation. The results showed that 1% variation of threshold value resulted in a 3.4% variation in BV/TV, 2% in Tb.N, 3.1% in Tb.Th, 2.9% in BS/BV, 1.8% in Tb.Sp, 29.2% in Exx, 28.7% Eyy and 27.7% in Ezz. Statistical analysis showed that 2.9% threshold variation caused significant change in BV/TV, Tb.Th, Exx, Eyy and Ezz values. The study shows that with threshold variation of more than 2.9%, significant differences in the architectural parameters and stiffness compared to those based on Otsu's method.