In order to analyze and evaluate the spectrum restoration property of image restoration algorithms, a new method based on Gaussian function is proposed. The optical transfer function H and degraded image spectrum function G are assumed as Gaussian functions, and the variance and variance ratio are used as indexes of the spectrum width to analyze and evaluate the spectrum restoration property of the image restoration algorithms quantificationally. The curves of H and G are enacted in two groups of different variances, and the variance and variance ratio of the image spectrum restored by constrained least squares restoration (CLS) and maximum likelihood estimation (PML) are calculated in no-noise and with-noise cases in analysis. The two algorithms are analyzed and evaluated by the calculation results, and the analysis proves that PML is better than CLS in spectrum extrapolation and noise restraint. The experimental results show that the method is easy and effective.