摘要： Wavelength selection is a critical step in multivariate calibration. Variable selection methods are used to find the most relevant variables, leading to improved prediction accuracy, while simplifying both the built models and their interpretation. In addition, different spectrophotometer designs and measurement principles result in non-destructive technologies applied in many fields, such as agriculture, food chemistry and pharmaceutics. However, an on-chip or portable device does not allow acquiring data from a large number of wavelengths. Therefore, the most informative combination of a limited number of variables should be selected. The Replacement Orthogonal Wavelengths Selection (ROWS) method is described here as a new method. This algorithm aims at selecting as few wavelengths as possible, while keeping or improving the prediction performance of the model, compared to when no variable selection is applied. The ROWS is applied to several near infrared spectroscopic data sets leading to improved analytical figures of merits upon wavelength selection in comparison to a built PLS model using entire spectral range. The performance of the ROWS-MLR method was compared to the FCAM-PLS method. The resulting models are not significantly different from those of FCAM-PLS; however, it involves a significantly smaller amount of variables.