This research article introduces a method to systematically assess which network generation algorithms best describe biological networks. The approach involves mapping graphs to a high-dimensional "word space" based on subgraph features, enabling the construction of classifiers that can distinguish between different network models. The method uses support vector machines (SVMs) to classify networks and evaluates the performance of these classifiers using empirical test loss. The study includes 17 different network generation models, including those inspired by biological processes, and applies the method to three real biological networks: the E. coli genetic network, the C. elegans neuronal network, and the S. cerevisiae protein interaction network. The results show that duplication-mutation schemes are most successful in describing these networks, although some models based on preferential attachment also perform well. The study highlights the importance of considering multiple features and substructures when evaluating network models, and demonstrates that the proposed method can effectively distinguish between different network generation mechanisms. The method is a first step towards systematizing network models and assessing their predictive power, and has potential applications in various scientific communities. The study also includes detailed descriptions of the network models, the word space, and the SVMs used for classification. The results are supported by statistical analysis, including kernel density estimation and robustness measures, which help to interpret the classification outcomes. The study concludes that the proposed method provides a powerful tool for analyzing biological networks and understanding their underlying structures.This research article introduces a method to systematically assess which network generation algorithms best describe biological networks. The approach involves mapping graphs to a high-dimensional "word space" based on subgraph features, enabling the construction of classifiers that can distinguish between different network models. The method uses support vector machines (SVMs) to classify networks and evaluates the performance of these classifiers using empirical test loss. The study includes 17 different network generation models, including those inspired by biological processes, and applies the method to three real biological networks: the E. coli genetic network, the C. elegans neuronal network, and the S. cerevisiae protein interaction network. The results show that duplication-mutation schemes are most successful in describing these networks, although some models based on preferential attachment also perform well. The study highlights the importance of considering multiple features and substructures when evaluating network models, and demonstrates that the proposed method can effectively distinguish between different network generation mechanisms. The method is a first step towards systematizing network models and assessing their predictive power, and has potential applications in various scientific communities. The study also includes detailed descriptions of the network models, the word space, and the SVMs used for classification. The results are supported by statistical analysis, including kernel density estimation and robustness measures, which help to interpret the classification outcomes. The study concludes that the proposed method provides a powerful tool for analyzing biological networks and understanding their underlying structures.