Fluoride and chloride reciprocal salts are potential novel media with suitable working temperature and high latent heat for next-generation solar power. A back propagation (BP) artificial neural network (ANN) algorithm was developed based on the known data of salts. The composition and melting point of two unknown binary fluoride and chloride reciprocal salts were predicted by the trained ANN model. The predicted composition and melting point of the reciprocal salts were verified by experimental tests. The predicted results of composition are in good agreement with the experimental values, and the predicted errors of the melting point are less than 1.5%. The melting point and fusion enthalpy of KCl-NaF reciprocal eutectic salt are 648 +/- 2 degrees C and 365 +/- 5 J/g, respectively. The thermal stability of this reciprocal eutectic salt is very good and the weight loss is still less than 3.0% even up to 800 degrees C. The good performance of KCl-NaF reciprocal eutectic salt at high temperatures suggest that it can be a good candidate for thermal energy storage systems with supercritical CO2 cycles. The ANN is an effective method to prediction composition and properties of molten salts, this method is expected to a quick method for design and selection of phase change material for the high temperature latent heat energy storage systems.