The paper presents a novel method for extracting optimal interatomic potentials from large datasets generated by first-principles calculations. This method, called the force-matching method, fits the potential to atomic forces of various configurations, including surfaces, clusters, liquids, and crystals at finite temperatures. The approach overcomes the limitations of traditional fitting methods by using extensive data, allowing for potentials with high accuracy comparable to *ab initio* methods. The method involves minimizing an objective function that combines forces from first-principles calculations and additional constraints on physical properties. The authors demonstrate the effectiveness of this method by applying it to aluminum, achieving a root mean square deviation of 0.17 eV/Å in force components and showing good agreement with experimental data in surface energies, surface relaxations, thermal expansion, and melting temperature. The study highlights the potential of the force-matching method for constructing realistic classical potentials with high transferability for systems that can be treated by *ab initio* methods.The paper presents a novel method for extracting optimal interatomic potentials from large datasets generated by first-principles calculations. This method, called the force-matching method, fits the potential to atomic forces of various configurations, including surfaces, clusters, liquids, and crystals at finite temperatures. The approach overcomes the limitations of traditional fitting methods by using extensive data, allowing for potentials with high accuracy comparable to *ab initio* methods. The method involves minimizing an objective function that combines forces from first-principles calculations and additional constraints on physical properties. The authors demonstrate the effectiveness of this method by applying it to aluminum, achieving a root mean square deviation of 0.17 eV/Å in force components and showing good agreement with experimental data in surface energies, surface relaxations, thermal expansion, and melting temperature. The study highlights the potential of the force-matching method for constructing realistic classical potentials with high transferability for systems that can be treated by *ab initio* methods.