Agriculture production largely depends on weather conditions and is extremely prone to natural hazards. A more frequent and severe occurrence of natural hazards such as storms and floods has put food security at increased risk in recent decades. Evaluating the true impact (loss and damage) of disaster in the agriculture sector is very challenging. The present study focusses on using a zrandomized field experimental approach at both district and micro agricultural-plot levels to investigate the impact of floods on agricultural yields in Sri Lanka and its effect on farmers who are averse to taking risks and those who are willing to take risks. A detailed site selection technique has been used in the study. The dissimilarity in difference estimates indicates that flood-affected households have experienced the loss of paddy and non-paddy crops. However, the net loss of non-paddy is higher than that in paddy. Farmers offset this loss by expanding crop cultivated areas zthat utilize soaked fields after the flood, though there are risks of pest attack and diseases. The results are not driven by household-specific characteristics and are robust to several specifications, different crop types and alternative flood-severity measures.