Infrared (IR) spectroscopic techniques combined with multivariate calibration (MVC) methods are promising for on-line monitoring. In a previous article [M. Zeaiter, M. Roger, V. Belon-Maurel, D. Rutledge, Trends Anal. Chem. 23 (2004) 157], robustness of the calibration was defined and different ways to evaluate it were identified. In order to improve the robustness of these calibration methods for industrial applications, an overview is presented of the existing methods, usually used to enhance prediction-model performance. The first part focuses on geometric spectral pre-processing methods, such as normalization methods, smoothing and derivatives. The second part discusses dimensionalityreduction methods, represented by orthogonalization and variable- selection methods. The impact of each method on the enhancement of the robustness of models developed by MVC is analyzed and discussed.