For non-invasive measurement of human blood cholesterol concentration, this experiment was carried out on 80 volunteers clinically. In vivo dynamic spectra of fingers were achieved and biochemical examinations of blood components contents including cholesterol were get as soon as possible. BP artificial neural network with inputs of dynamic spectra plus energy of harmonic waves processed by Principal Components Analysis(PCA) was used to establish the model of the total cholesterol values. The correlation between the predicted value and the true value of cholesterol is 96.48%. The maximum relative error is 25.44% and root-mean-square error of prediction is 0.242 6 mmol·L-1. The results show that PCA can make the process of computing faster and this study is another advance of dynamic spectra.