As the first step of the research project in assessing and accounting for model related errors in climate forecasting, an experiment of ensemble forecasts with multiple versions of the Global Forecasting System (GFS) model was carried out at National Centers for Environmental Prediction (NCEP). Two versions of the model were used for ensemble forecasts up to 16 days, one with the Simplified Arakawa-Schubert (SAS) scheme, and the other with the Relaxed Arakawa-Schubert (RAS) scheme as the Convective Parameterization Scheme. A third version of the model, with the Convective Parameterization Scheme neglected, was also integrated as a benchmark. The performance of the ensemble forecasts, including probabilistic forecasts, for the two month period of August and September 2002, from the three original ensembles and their combinations, are analyzed and compared, with an emphasis on the 500hPa height and the precipitation. The results of the analysis will be presented and their implications for further investigations into the model-related errors in ensemble forecasting will be discussed.