The physical origin of the negative thermal correction to the Casimir force between metals is clarified. For this purpose the asymptotic behavior of the thermal Casimir force is analyzed at large and small distances in the real frequency representation. Contributions from propagating and evanescent waves are considered separately. At large distances they cancel each other in substantial degree so that only the attractive Lifshitz limit survives. At smaller separations the repulsive evanescent contribution of s-polarization dominates in the case of two metals or a metal and a high-permittivity dielectric. Common origin and order of magnitude of the repulsion in these two cases demonstrate naturalness of the controversial large thermal correction between metals. DOI: 10.1103/PhysRevA.76.062102 PACS number͑s͒: 12.20.Ds, 42.50.Lc, 42.50.Pq Over the last decade there was a growing interest in the Casimir force ͓1͔ that was measured in a series of recent experiments ͓2͔. Particularly the thermal part of the force was a subject of active and often controversial discussion ͑see ͓3͔ for a review͒. Here we concentrate on the thermal force, which does not include the zero point fluctuations. The repulsion discussed in this paper has the meaning of a negative thermal correction to the force at zero temperature, but the total force is always attractive ͓4͔. At large distances a ӷបc / T ͑k B =1͒, the thermal force is given by the Lifshitz limit ͓4-6͔where a is the distance between parallel plates and T is the temperature of the system. This formula was derived for two dielectric plates with the static dielectric constants 1 and 2 . The force between two ideal metals can be found from Eq. ͑1͒ as the limit 1,2 → ϱ that gives F Lif = T ͑3͒ / 8 a 3 . This equation became one of the points of controversy ͓7,8͔ because direct calculation of the thermal force between ideal metals ͓9͔ gave the result, which is two times larger. This contradiction found its resolution ͓10,11͔. At large distances only low frequency electromagnetic ͑EM͒ fluctuations contribute to the force. In this limit the s-polarized EM field degenerates to a pure magnetic one, which penetrates freely via a nonmagnetic metal ͓11͔. On the contrary, the ppolarized field is pure electric and reflected by the metal. For the ideal metal both polarizations are reflected and the force will be two times larger. In this sense the ideal metal is rather the limit case of a superconductor than of a normal metal ͓12͔.The difference between ideal and real metals manifests itself also at small distances a បc / T. The force between ideal metals is attractive and small ͓9͔. On the contrary, the force between real metals is relatively large and repulsive ͓7͔. This difference did not yet find a clear physical explanation.In this paper it is demonstrated that at distances a Շបc / T the evanescent contribution of s-polarization to the thermal Casimir force dominates for two metals or in the case of a metal and a high-permittivity dielectric. For both material configurations the force...