The paper describes the production of a high spatial resolution (30 m) soil type map for the pan-EU based on the IUSS's World Reference Base classification system and Ensemble Machine Learning with a large set of covariates. 19,680 legacy soil survey points collated from multiple national and European projects were harmonized and combined to produce a consistent set of training points with a total of 185 classes. The training points were overlaid by 130 covariate layers containing climatic variables (CHELSA Bioclimatic variables), Landsat-based long-term (2000-2022) biophysical indices, multiscale terrain variables (at 30, 60, 120, 240, 480 and 960 m resolutions) and parent material variables. Predictions of soil-type probabilities were produced by fitting an ensemble model based on Random Forest and LightGBM. The results of the variable importance analysis indicate that the model is mainly driven by relief and climate factors. The accuracy assessment reveals a moderate decrease in the logarithmic loss of the model from 4.41 to 2.76 (about 37 %) compared to the dummy model. Visually, the produced predictions match existing coarse resolution (1 km) pan-EU products, but then reveal significantly more detail, especially in areas with complex topography. The intended uses of the 30 m resolution maps include: (1) to help distinguish different soil regions (cross between soil types and climatic variables), (2) to assist in predictive soil mapping of soil properties with limited laboratory data, (3) to serve as a covariate in agronomy and forest species distribution models, e.g. to map yields and for species distribution mapping. The data and code used are publicly available under an open license from https://doi.org/10.5281/zenodo.13838408 and https://github.com/AI4SoilHealth/.