The three-dimensional (3-D) modelling of the spinal shape by least square Fourier series to evaluate local measurement of geometric torsion for scoliotic spine shows limitations. In fact, this method does not allow the definition of a “true” 3-D inflexion point in double major scoliotic curves. Torsion values computed in 3-D inflexion regions are consequently non-representative since they result in numerically instable “large torsion spike”. Simulation were performed to analyzed the torsion “spike” problem and revealed that numerical instabilities occur because the second derivative of parametric functions x(t), y(t) and z(t) does not equal zero for the same t value. The development of a “torsion spike” elimination technique, by the imposition of a “true” 3-D inflexion point, results in numerical stability, without significantly modifying the general shape of the scoliotic curve. Applying Frenet’s formulas to this resulting corrected curve allow the analysis of the geometric torsion phenomenon associated with many types of scoliotic shapes.