CityNav is a new dataset designed for language-goal aerial navigation, leveraging 3D point cloud data from real-world cities. The dataset includes 32,637 natural language descriptions paired with human demonstration trajectories, collected through a web-based 3D simulator. Each description specifies a navigation goal, incorporating landmark names and locations. The dataset aims to bridge the gap in aerial navigation research by providing a comprehensive resource for training and benchmarking aerial navigation agents. Key findings from the dataset include the superior performance of models trained on human demonstration trajectories over those trained on shortest path trajectories, and the significant enhancement of navigation efficiency when integrating a 2D spatial map. The dataset and code are available online, making it a valuable resource for advancing aerial navigation systems.CityNav is a new dataset designed for language-goal aerial navigation, leveraging 3D point cloud data from real-world cities. The dataset includes 32,637 natural language descriptions paired with human demonstration trajectories, collected through a web-based 3D simulator. Each description specifies a navigation goal, incorporating landmark names and locations. The dataset aims to bridge the gap in aerial navigation research by providing a comprehensive resource for training and benchmarking aerial navigation agents. Key findings from the dataset include the superior performance of models trained on human demonstration trajectories over those trained on shortest path trajectories, and the significant enhancement of navigation efficiency when integrating a 2D spatial map. The dataset and code are available online, making it a valuable resource for advancing aerial navigation systems.