The prices of walnuts vary according to their geographical origin and, therefore, offer a financial incentive for adulteration. A reliable analysis method is required to quickly detect possible misdeclarations and thus prevent food fraud. In this study, a method to distinguish between seven geographical origins of walnuts using Fourier transform near-infrared (FT-NIR) spectroscopy combined with chemometrics as a fast, versatile, and easy to handle analytical tool was developed. NIR spectra of 212 ground and afterwards freeze-dried walnut samples, harvested in three consecutive years (2017–2019), were collected. We optimized the data pre-processing by applying and evaluating 50,545 different pre-processing combinations, followed by linear discriminant analysis (LDA) which was confirmed by nested cross-validation. The results show that in the scope of our research minimal pre-processing led to the best results: By applying just multiplicative scatter correction (MSC) and median centering, a classification accuracy of 77.00% ± 1.60% was achieved. Consequently, this complex model can be used to answer economically relevant questions e.g., to distinguish between European and Chinese walnuts. Furthermore, the great influence of the applied pre-processing methods, e.g., the selected wavenumber range, on the achieved classification accuracy is shown which underlines the importance of optimization of the pre-processing strategy.