This article uses the partial least square algorithm and the improved differential evolution algorithm to preprocess the ultraviolet-visible absorption spectral data and build a model for detection of cholesterol concentration in human serum. The result indicates that the wavelet decomposition and the partial least square algorithm can effectively reduce the multiple correlation coefficients. The absorption spectrum is investigated by different basis functions with various pre-processed spectra, and produces the best result with the prediction error of 0.12mmol/L. This study shows that the ultraviolet-visible absorption spectroscopy can offer a feasible research direction for detection and quantitative analysis of cholesterol concentration.