The stellar continuum is a sort of spectrum whose light intensity changes continuously and smoothly with wavelength (frequency) due to blackbody radiation. Each observed spectrum contains continuous spectra, spectral lines and noises. The classification of stellar is mainly based on the spectral lines of the spectrum, relative intensity of the continuum and other characteristics of the spectrum. The distribution of the stellar continuum and the contour of the lines are determined by the stellar atmospheric parameter, so the stellar atmospheric parameter can be estimated from the continuum and the spectral lines. Therefore, the main problem of the spectral data processing is to extract the continuum and the lines. The current algorithms for stellar continuous spectral extraction are mainly polynomial approximation, median filtering, morphological filtering and wavelet filtering. However, these methods for the robustness of low-quality spectral processing are not very satisfying. Therefore, it is necessary to study a new algorithm for extracting the continuous spectrum from the low-quality spectra. In this paper, a fitting method for low-quality stellar spectrum based on Monte Carlo is proposed after careful analyses of low-quality stellar continuum. The method is used to automatically interpolate at the point where the spectrum is not in the range of the star spectrum with Monte Carlo, so each wavelength corresponds to a flow point, and then the low-order polynomial iterations are fitted to these flow points for obtaining the continuous spectrum. In order to verify the robustness of the algorithm for low-quality spectral continuum extraction with different SNRs, we use different SNRs to simulate different low-quality spectrum by adding different Gaussian white noise to the original spectrum. The result shows that the proposed algorithm has high accuracy and robustness to the fitting of low-quality spectrum with different SNRs.