Abstract: Terahertz multispectral imaging is an emerging technique for material identification, as many molecules present uniquespectral signatures in the frequency range that typically spreads from 100 GHz up to several THz. In most cases, samples to identify are imaged in transmission, which is most of the time not the appropriate configuration regarding industrial requirements. Unfortunately, in reflection configuration, the sample thickness leads to terahertz beam misalignment (defocusing, lateral shift…) that involves erroneous estimation of the reflected amplitude. Such misalignment prevents any absolute characterization of the sample under testing.Thus, to evaluate libraries of materials, one must prepare samples showing exactly the same thickness. This process could be extremely time-consuming and very often, not applicable for fast evaluation in the industry. In this paper, we propose a novel method that evaluates the effect of the unwanted misalignment artifacts, due to the varying sample thickness. Such a method is applied to correct the THz amplitude image and to get an absolute image in reflection configuration, in order to compare library of materials whatever the sample thickness.