Abstract: In order to evaluate the size of image noise of electron multiplying CCD (EMCCD) quantitatively and achieve an accurate estimation of the EMCCD image noise parameters, the ECMMD noise distribution and its parameter estimation methods were studied. Firstly, the sources and statistical properties of EMCCD noise were discussed, and thereby an EMCCD noise distribution model was established. Then two EMCCD noise parameter estimation methods --the moment estimation method and the Gauss -Newton method were proposed and Monte Carlo simulation was done to verify their performance. The results show that both the average relative error and the relative standard deviation of the two methods are of 10-2 magnitude, presenting high estimation accuracy, and the Gauss -Newton method get better performance. With integration time of 50 s, a series of EMCCD images of the dark field with no gain and images of the background with gain of 50 were obtained. Using the moment estimation method and the Gauss-Newton method, the dark current noise, clock induced charge noise and readout noise were estimated. The results present that the estimated value is consistent with the EMCCD index value, which proves the moment estimation method and the Gauss -Newton method are able to estimate the noise parameter effectively and has a high accuracy.