This paper introduces FGVC-Aircraft, a new dataset containing 10,000 images of aircraft models, organized in a three-level hierarchy: variant, family, and manufacturer. The dataset aims to study fine-grained visual classification of aircraft, which presents unique challenges due to the subtle but visually measurable differences between models. The dataset was constructed using online resources and contributions from aircraft enthusiasts, a strategy that can be applied to other object classes. The paper defines three classification tasks—variant, family, and manufacturer recognition—and presents baseline results. The dataset is publicly available for research purposes, with detailed annotations and evaluation protocols. The authors highlight the potential of aircraft recognition as a novel domain in fine-grained visual classification and plan to expand the dataset by including more models and applying similar methods to other object categories.This paper introduces FGVC-Aircraft, a new dataset containing 10,000 images of aircraft models, organized in a three-level hierarchy: variant, family, and manufacturer. The dataset aims to study fine-grained visual classification of aircraft, which presents unique challenges due to the subtle but visually measurable differences between models. The dataset was constructed using online resources and contributions from aircraft enthusiasts, a strategy that can be applied to other object classes. The paper defines three classification tasks—variant, family, and manufacturer recognition—and presents baseline results. The dataset is publicly available for research purposes, with detailed annotations and evaluation protocols. The authors highlight the potential of aircraft recognition as a novel domain in fine-grained visual classification and plan to expand the dataset by including more models and applying similar methods to other object categories.