We discuss mean field theory of glasses without quenched disorder focusing on the justification of the replica approach to thermodynamics. We emphasize the assumptions implicit in this method and discuss how they can be verified. The formalism is applied to the long range Ising model with orthogonal coupling matrix. We find the one step replica-symmetry breaking solution and show that it is stable in the intermediate temperature range that includes the glass state but excludes very low temperatures. At very low temperatures this solution becomes unstable and this approach fails.The thermodynamics of glasses without quenched disorder is a long standing problem in statistical physics. The interest to this problem was renewed recently when it was understood that powerful methods developed for the glasses with quenched disorder can be often applied to this problem. [1][2][3][4][5][6]. In both systems the local magnetization in the ground state varies from site to site and different sites are typically non-equivalent. The qualitative reason why glasses without quenched disorder are more difficult to describe theoretically than spin glasses is the following. The mean field theory has to operate with the average magnetization (or its moments), not with quantities which depend on a realization and a particular state. The average quantities appear naturally in spin glasses after averaging over quenched disorder which makes all sites equivalent.A few methods were suggested to overcome this difficulty for the glasses without quenched disorder. First, a mapping of some glass models to the quenched disordered problems was suggested [1], this method has an obvious disadvantage that such mapping is difficult to guess. Second, it was noted that a typical dynamics in a glassy system leads not to a ground state but to one of many metastable states providing an effective averaging mechanism [7] which makes all sites equivalent even for glasses without quenched disorder. This method has a disadvantage that dynamical equations are much more difficult to solve than the statical ones. Very recently the cloning method was proposed that is based on the idea that even at low T a system of m clones might be distributed in its phase space over many low lying metastable states if m is chosen correctly and the properties of all these states are essentially equivalent to those of the ground state. [3][4][5][6] Generally, the partition sum of m weakly coupled clones is F e −N (mβF −S conf (F )) , where sum goes over free energies (per site) of metastable states, F , and S conf (F ) is their configurational entropy (S conf = 1 N ln(N states )). Assuming that dS conf (F )/dF is finite at the lowest F associated with the ground state one needs to chose m ∝ T at low T in order to avoid a complete dominance by a single (ground) state and the problems with site nonequivalence mentioned above. Distributing the system in the phase space provides the effective averaging mechanism in this approach. The main assumptions implicit in this approach are that...