Multi-model or multi-model version ensembles resulted in various results and the effects of model diversity and analysis diversity remain to be understood. Cumulus parameterization has been viewed as one of the most important components of global circulation models, yet the variations associated with the use of different schemes or no such scheme at all, is still not clear. On the other hand, the effect of variation in the structure and intensity of horizontal diffusion on the characteristics of ensemble forecast has not been systematically studied. In this paper, these issues are investigated following a single model, multi-version approach, using NCEP's GFS model. Systematic and random variations in model output associated with variation in the model used are examined by verifying both ensemble and single deterministic forecasts. For cumulus parameterization, two well-tested schemes are used and a third model version is formed by switching it off. With the horizontal diffusion, its order and coefficient are systematically altered to form different versions of the model. The results of experiments suggest that the systematic variation associated with that in horizontal diffusion is more significant than that in cumulus parameterization, for atmospheric flow variables over the extratropics. The implication of the results in the generation of multi-model or multi-version ensembles in global ensemble forecasting, is also discussed.