The paper summarizes the developments of the National Centers for Environmental Prediction (NCEP), the European Centre for Medium-Range Weather Forecasts(ECMWF), and the Canadian Meteorological Centre(CMC), which are the most representative of global ensemble forecast system(GEFS).Due to the enlarging of computational resources, the model resolution and ensemble size of their GEFS subsequently increase.At the same time, for promoting the improvement of the forecast skill, they all devote to develop the initial and model perturbation methods used to simulate the effect of initial and model uncertainties.The initial perturbation methods are updated from the singular vector (SV) method(ECMWF), the breeding method (NCEP) and the perturbed observation (PO) method (CMC) to the ensemble of data assimilation and singular vector(EDA-SV) method(ECMWF), the ensemble transform with rescaling(ETR) method(NCEP) and the ensemble Kalman filter(EnKF) method(CMC).Several attempts are also made to account for model-related uncertainty.ECMWF and CMC have revised their stochastic physics parameterization tendencies(ECMWF) and multi-parameterization(CMC) schemes, and NCEP also develops stochastic total tendency perturbation to estimate the model-related uncertainty.To accelerate improvements on the accuracy of global high-impact weather forecasts, TIGGE(the THORPEX interactive grand global ensemble) was initiated to enhance international collaboration on multi-center and multi-model ensemble forecast, and NAEFS (North American ensemble fo recast system) can provide an operational framework for global multi-model ensemble forecast system.They are all helpful for developing the global interactive forecast system(GIFS).