Abstract: Based on the discussion of image noise sources and their statistic characteristics of the electron multiplying CCD(EMCCD), the Poisson-Gaussian-mixture noise distribution model was established. Aiming at the problem that the solution of the maximum likelihood function of the Poisson-Gaussian-mixture distribution model was difficult to solve, the expectation-maximization method was proposed to estimate the parameters of Poisson-Gaussian-mixture noise distribution model of the EMCCD after appropriate initialization settings on the noise model, reducing the complexity of the parameter estimation and achieving equivalent effect of the maximum likelihood estimation. Monte Carlo simulation results and experimental results show that the expectation-maximization method can achieve good performance, provide satisfied fitting features for Poisson-Gaussian-mixture distribution, and obtain high precision parameter estimation values.