Growth theory predicts that natural disasters should, on impact, lower GDP per capita. However, the empirical literature does not offer conclusive evidence. Most existing studies use disaster data drawn from damage records of insurance companies. We argue that this may lead to estimation bias as damage data and the selection into the database may correlate with GDP. We build a comprehensive database of disaster events and their intensities from primary geophysical and meteorological information. In contrast to insurance data, our GeoMet data reveal a substantial negative and robust average impact effect of disasters on growth. The worst 5% disaster years come with a growth damage of at least 0.46 percentage points. That average effect is driven mainly by very large earthquakes and some meteorological disasters. Poor countries are more strongly affected by geophysical disasters; rich more by meteorological events. International openness and democratic institutions reduce the adverse effect of disasters.