Trabecular bone specimens were tested by non-destructive technique with the purpose of investigating stiffness behaviour and optimizing stiffness determination. Cylindrical specimens (n = 25) were loaded repetitively (0.1 Hz, 30 cycles) by axial compression to 50% of predicted ultimate strength and finally compressed to failure. Analyses of single compression curves showed increasing stiffness (E') until a stress level about 50% of ultimate stress followed by decreasing stiffness. Curve fit analysis of the elastic part of the compression curve showed the best fit, when a second order polynomial was used (r = 0.94, p less than 0.001). The stiffness determined non-destructively at the 25% level of ultimate strength increased significantly to the tenth loading cycle followed by a steady state. The precision of stiffness determination as an average of five consecutive measurements at steady state was E' +/- less than 5% (95% confidence limits). A reproducibility test by repetition of the test sequence after 3 h rest showed qualitatively the same stiffness behaviour. The variation of stiffness determination between the two test sequences was +/- 27% at the first loading cycle falling to +/- 12% at steady state.