We proposed a new definition of drug-likeness based on protein-compound docking simulation. Active and decoy compounds of 40 target proteins were investigated. These compounds were docked to protein sets consisting of 53-160 proteins. The protein sets did not include the target proteins. The average value and deviation of docking scores against the multiple proteins were calculated for each compound. Our study revealed that the docking scores of active compounds are more widely distributed than those of decoy compounds. Thus, the deviation of docking scores with multiple proteins should be a measure of drug-likeness for compound affinity.