Abstract: The ellipticity, transmittance and magnetic susceptibility of diamond are taken as input and the impact toughness (TI) and thermal impact toughness (TTI) as output to establish the mapping relationship between the inputs and outputs by BP neural network. Then a BP neural network prediction on TI and TTI values of diamonds are obtained. The results show that the prediction of TI and TTI values are relative accurate. The average relative error rate between the predicted values and the detected ones is not higher than 1.4%, and the maximum relative error rate is not higher than 5.4%. Therefore, the BP neural network prediction could replace the current damage detection methods to some extent.