The paper introduces a specific non-destructive technique in use to detect the occurrence of geomaterial instability subsequently causing damage, and a specific data reduction procedure to assess damage accumulation. The damaging process was detected by analysing the signal evolution of ultrasonic pulses propagating through some sandstone specimens subject to recurrent freeze-thaw cycles under constant static loading. It is assumed that the evolution of non-linear energy increase corresponds to the process of the damage extent. An input-output non-parametric technique and a non-linear analyser for data reduction procedure were chosen to portray the non-linear aspects of ultrasound propagation characteristics. Based on a multi-dimensional Fourier transform, the non-linear analyser permits to separate linear and non-linear parts. It can be used to monitor non-destructively and continuously the overall alteration or damage process of a brittle rock specimen at initial damage caused by freezing and thawing, so that damage mechanism could be quantitatively estimated by a dimensionless parameter, the so-called non-linearity ratio. The experimental results showed that rock frost damage in the form of crack sets is directly related to the increase of non-linearity ratio. It reveals early microstructural damages that appear before occurrence of macroscopic rupture.