A baseline fitting algorithm for spectral signal region is proposed in order to eliminate the adverse effects of baseline distortion in spectral processing. First， a GradSuck fitting algorithm is proposed for baseline with abrupt curvature by using the idea of gradient inertia force and suction， which cover the shortage of the piecewise quadratic polynomial fitting algorithm when the baseline curvature around the signal area changes suddenly. Then this algorithm is combined with the piecewise quadratic polynomial fitting algorithm to propose a more general fitting algorithm for spectral signal region. At the same time， the proposed algorithm has been compared with different baselines fitting algorithms. Experiments under various baseline types with different SNR show that the baseline fitting algorithm of signal region has reliable accuracy and stability， and it can extract the spectral baseline better than other algorithms. The relative error of its whole fitting accuracy is only 47.0% of the quadratic polynomial fitting， 35.6% of the AirPLS fitting， and 20% of the wavelet fitting. And it also has high real-time performance because of only fitting the baseline of the signal region.