A noise-component removing method for removing a noise component from multipoint spectral data that has been generated through measurements performed at measurement points of a sample surface, the method comprising: a PLS analysis step of determining components of the multipoint spectral data for each measurement point in a descending order of eigenvalues of the components by subjecting the multipoint spectral data to multivariate analysis based on the partial least squares regression using a value obtained by quantifying characteristic information about a characteristic of each measurement point, other than spectral information of the measurement point and using the spectral information as an independent variable in the partial least squares regression; and a spectrum reconstruction step of reconstructing the multipoint spectral data for each measurement point to eliminate a component having an eigenvalue lower than a predetermined value, from the components determined in the PLS analysis step.