Outlier samples often strongly influence the stability of the model in ultraviolet-visible absorption spectroscopy.Eleven outliers were detected and eliminated from 108 Gastrodia elata blume(GEB) samples by the method of PCA-mahalanobis distance.The population was divided into calibration and validation sets in order to perform an external validation.For that purpose,one of each six samples was selected to build up the validation set,and the rest remained as the calibration set.Calibration from the resulting set of Savitzky-Golay smoothing and differentiation was built up a model using simplified partial least squares regression.By extending the number of smoothing point to 3,5,7,…,25,(odd) and polynomial degree to 2,3,4,5,6,including 162 methods from Savitzky-Golay smoothing and differentiation were calculated.The optimal model was selected by the prediction effect.And the derivation order was 2,the polynomial degree was 4,the number of smoothing points was 5,the optimal correlation coefficient(r) reached 0.910 2,the optimal root mean standard of prediction(SEP) reached 0.390 7 g/L.The results indicated that the using of PCA-mahalanobis distance,Savitzky-Golay smoothing and differentiation and SIMPLS,can be effectively applied in the model optimization of ultraviolet-visible spectroscopy in determining Gastrodin in GEB.