This Phd thesis focuses on the generation and optimization of calibration models. When calculating the regression model based on the Principal Component Regression (PCR), the selection of factors for calibration represents the most important and time consuming step. During this step relevant spectral information has to be separated from interfering artefacts, as e.g. noise, minority effects, etc. This is achieved by Genetic Algorithms (GAs), which are designed to find good solutions for this combinatoric optimization problem by making use of adaptive processes. Based on examples taken from typical application fields of NIR spectroscopy, the evaluation and optimization of calibrations is described in detail. The GA serves to calculate the most suitable factor combination in a reproducable and reliable way. Therefore the description and validation of the GA's functional design represents a further main topic.