The calculation of vegetation metrics from LiDAR point clouds might be affected by the available point density of a dataset. Testing how the same LiDAR vegetation metrics differ with different point densities can therefore inform about their robustness for upscaling metrics to other areas or other LiDAR point clouds. The datasets made available here were generated to test the robustness of LiDAR vegetation metrics to varying point densities. A total of 25 LiDAR vegetation metrics representing different aspects of vegetation height, vegetation cover and structural complexity were tested (see metric definition in Kissling et al. 2023, https://doi.org/10.1016/j.dib.2022.108798). The metric calculation was similar to the metric calculation in the Laserchicken software (Meijer et al. 2020, https://doi.org/10.1016/j.softx.2020.100626) and the Laserfarm workflow (Kissling et al. 2022, https://doi.org/10.1016/j.ecoinf.2022.101836). The Dutch AHN4 dataset from the years 2020–2022 with a point density of 20–30 points/m2 was used. A number of plots (i.e., squared polygons around centre points) were randomly placed across the Netherlands within Dutch Natura 2000 sites (using shapefiles from the European Environmental Agency). Different Dutch Natura 2000 sites were distinguished based on their dominant habitat type (dunes, grassland, marsh, shrubland, and woodland). About 100 plots were randomly placed in each habitat type. The AHN4 point cloud of each plot was clipped and then randomly downsampled to 1, 2, 5, 10, 15, 20 points per square meter, respectively. This was done for six different spatial resolutions (1, 2, 5, 10, 20 and 30 meter). The clipped points were then used to calculate the 25 LiDAR vegetation metrics for the original point density and for the six down-sampled point densities.