The relationships between thermal conductivities and soluble solid contents, water contents, temperature, density and hardness of 30 types fruits and vegetables were studied by the tepid probe test system. A BP artificial neural network model was presented and optimized for the prediction of thermal conductivities of fruits and vegetables according to the error analysis. The result showed that the optimal model was able to predict thermal conductivity with a mean relative error of 1.11%, a mean absolute error of 0.0057W/(m?K). The model can be incorporated in heat transfer calculations during fruits and vegetables processing. 