To determine multi-constituent of Traditional Chinese Medicine (TCM) by spectral analysis, a new method for establishing the calibration model of spectral data which is nonuniform distribution and nonlinear, named local fitting principal component regression (LFPCR), is proposed. The weight of eveiy sample in the known sample set is calculated according to the distance between the test set data and the known sample set data, and then a calibration set is composed of the samples with relatively greater weights. Based on the calibration set, the calibration model can be established by linear fitting of multi- section. Applying the proposed method to determining jtrorrhizine, plmatine and berberine of Coptis chinensis, the root mean square errors of prediction results obtained by LFPCR are 0.023, 0.040 and 0. 052 respectively, which are superior to those obtained by PCR, PLS and ANN. Therefore, the method can be used to determine multi-constituent of TCM at higher accuracy.

Key words: RHOTOMETRY, CHINESE DRUG, BERBERINE, ALKALOID, SPECTROPHOTOMETRY