A dam-break flood can cause catastrophic life losses due to the uncontrolled large amount of water. This paper aims to improve the estimation of loss of life (LOL) from dam failure in China. For this purpose, we collect fifteen dam failure cases in China. Each failure case is further divided into subcases, if possible, based on flood severity and distance between the flooded area and the dam site. As a result, there are totally twenty-two subcases. Using the subcases as the basic data, an improved method is proposed for determining LOL that occurred from dam break by dividing the flood severity into two conditions: low flood severity, and medium and high flood severity. Correspondingly, two empirical equations are developed based on the multi-variable regression analysis for low flood severity, and medium and high flood severity, respectively. The proposed equations can predict a range for fatality rates from 0.0 to 1.0, and thus it can be applied for any case with any fatality rate. A comparison between the proposed equations and other methods is carried out, which shows that the proposed equations are much closer to the actual values with a coefficient of determination (R²) equals 0.9848, and 0.9872 for low flood severity equation and medium and high flood severity equation, respectively. Furthermore, the proposed empirical equations are applied to predict the life losses of three cases, two cases of low flood severity and one of medium flood severity. The results show a good agreement between the estimated life losses and actual life losses.