Nationality fabric classification is a significant work to promote the protection work of fabric patterns and further reveal its unique connotation and inheritance rules in big data era. Thus, how to ascertain the feature representation of fabric patterns becomes a primary problem. This paper presents a high-level feature representation for fabric patterns for nationality classification, called FabricGene, which improves the semantic expression ability of the fabric pattern features. In fabric patterns, each FabricGene represents a complete abstract concept including the external shape and connotation characteristics. We evaluate the performance of FabricGenes and basic geometric primitives to illustrate the effectiveness of FabricGenes in nationality classification. Five widely used classification algorithms are applied to classify the fabric patterns by learning from training data with 12 groups of FabricGenes and 11 groups of basic geometric primitives respectively. The results demonstrate that the FabricGenes perform more effectively and stably in nationality classification than the basic geometric primitives. Namely, the FabricGenes can express the fabric patterns’ nationality features more accurately.