The present paper presents a partial least squares (PLS) regression coefficient matrix based wavelength selection algorithm. The regression coefficient was used as the criterion for the wavelength selection, and the Root-Mean-Squares Error of Cross-Validation (RMSECV) curve was referred to, to decrease the primary number of wavelength selected. Based on it, the uninformative wavelength can be eeleted through iteration steps, and the prediction accuracy of the model can also be improved. The algorithm was compared with the existing methods via a hydrogenation process Raman spectra data set, and the results indicated that the new algorithm can produce a more accurate and concise model than the existing ones.