The objective of this study was to elucidate the range of abilities of nonclinical safety assessment for predicting adverse drug reactions (ADRs) in humans. The dataset included 1256 ADRs with an incidence rate of 5% or more collected from 142 drugs approved in Japan from 2001 to 2010 (excluding anticancer agents and vaccines). Gastrointestinal, neurological and hepatobiliary ADRs were relatively common, followed by hematological, cutaneous, systemic and cardiovascular ADRs in the dataset. The analysis revealed that 48% of ADRs were predictable based on a comprehensive nonclinical safety assessment considering animal toxicity. Hematological and ocular ADRs, infection, and application site reactions showed a correlation of more than 70%, while musculoskeletal, respiratory and neurological ADRs showed a correlation of less than 30%. In addition to subjective patient perceptions, several laboratory parameters routinely monitored both in animals and humans showed a lower correlation, e.g., abnormalities in hepatobiliary and metabolic parameters, and blood pressure increase. Large molecule drugs showed lower correlation than small molecule drugs; ADRs were observed in various organs and consideration of pharmacological action did not significantly contribute to the prediction. It was also confirmed that the current standard of toxicology testing regarding dosing duration and dose level is adequate to detect concordant animal toxicity. This study collectively demonstrated a significant value of nonclinical safety assessment in predicting ADRs in humans. It also identified the subset of ADRs with poor predictability, highlighting the need for advanced testing that enables successful translation of animal toxicity to clinical settings with better accuracy and sensitivity.