A general Bayesian framework for model selection on random network models regarding their features is considered. The goal is to develop a principle Bayesian model selection approach to compare different fittable, not necessarily nested, models for inference on those network realisations. The criterion for random network models regarding the comparison is formulated via Bayes factors and penalizing using the mostwidely used loss functions. Parametrizations are different in different spaces. To overcome this problem we incorporate and encode different aspects of complexities in terms of observable spaces. Thus, given a range of values for a feature, network realisationsare extracted. The proposed principle approach is based on finding random network models, such that a reasonable trade off between the interested feature and the complexity of the model is preserved, avoiding overfitting problems.