Mechanization is one of the key ingredients for achieving high agricultural productivity. Despite its importance, there is currently no globally comprehensive information about countries’ agricultural mechanization. Here, we propose and demonstrate a machine learning approach, relying on a large, novel training dataset, to not only produce an up-to-date and comprehensive dataset of countries’ average agricultural mechanization, but also a global gridded map at ~ 5km resolution. Comparing our results to previously available data we find major improvements in accuracy, completeness, timeliness etc., and we notice that several countries are by now much more mechanized than reported so far. When investigating the association between mechanization and crop yield gaps we find a strong and robust link: For each 10 percentage point increase in mechanization, the associated crop yield gap decreases by 4 – 5 percentage points.