An analysis model for mixed multiplicative and additive noise of spectral signal is built, and an algorithm combining Wiener filtering and homomorphic filtering is proposed to denoise the spectral signal. Simulation results show that the proposed algorithm has better performance than moving average algorithm, least mean square algorithm and recursive least square algorithm. Experimental results indicate that the noise in xenon lamp spectral signal matches the mixed multiplicative and additive model. Compared with moving average algorithm, least mean square algorithm, and recursive least squares algorithm, after the mercury lamp spectral signal is processed by the proposed algorithm, more stable characteristic spectral parameters can be extracted, such as peak and valley locations, peak amplitude and full width at half maximum. Better result can be obtained in quantitative analysis with the proposed algorithm.