Raman spectra measured in the finger are a combination of backscattered signals induced from incident light. They contain Raman scattering, intrinsic tissue fluorescence and noise. Our goal of this study is to find perturbing components which are able to prohibit an accurate prediction of target materials and propose a new terminology for the SNR of regression coefficient vector. We assumed that Raman spectra are consist of fluorescent background spectrum (F), Raman spectrum of glucose (Rglucose), Raman spectrum of the skin (Rskin), random noise (RN) and unknown spectrum from other components (Runknown). And we used partial least squares regression (PLSR) to evaluate the main perturbing components under the condition that is various combinations of each signal. In simulation, Rskin is the main component to cause the error between reference and predicted glucose concentration. F is also able to affect accuracy at the low signal to noise ratio (SNR) of glucose signal. To minimize the perturbing effects, enhancement of Rglucose is more important than any other things.