. A free motion of a relativistic string is quantized by the use of Dirac's generalized canonical formalism. It is shown that only the light-like gauge is applied to quantization of this system and the result is the same as that of Goddard, Goldstone, Rebbi and Thorn which is obtained by a different method. The relativistic quantum mechanics is constructed in this gauge when the dimension of space-time is twenty-six and the intercept of the leading trajectory is unity. § I. Introduction Nowadays a massless relativistic string is considered to be a fundamental physi cal object of dual resonance models. A free motion of the string is beautifully formulated by Nambu and Gotol) in a covariant manner as a variational problem. The Lagr'!-ngian which is the starting point of their discussion about the string motion is a singular one in the sense that velocities cannot be solved as functions of canonical coordinates and their· conjugate momenta. Moreover, because the La grangian depends linearly on velocities/> the Hamiltonian function vanishes identi cally. Recently Goddard, Goldstone, Rebbi and Thorn (G. G. R. T.) discussed how to formulate the Hamiltonian fprmalism of the string model with the conclusion that the covariant quantization needs the dimension of space-time to be twenty-six and the intercept of the leading Regge trajectory to be unity, using the noncovariant gauge. 2> It seems, however, that there are some doubtful treatments in their dis cussion, which arises· mainly from identifying the displacement operator along the light-like direction with the new Hamiltonian instead of the original one, which is obtained from the Lagrangian and is numerically zero. To deal with this system properly from the viewpoint of the canonical quantiza tion formalism, we will adopt, in this paper, the generalized canonical formalism developed by Dirac3> and consistently eliminate redundant variables from the system described by a singular Lagrangian. As a result of this formalism, it is pointed out that the light-like gauge is favourable to quantize this system rather than the time like gauge. Our conclusion is against that of Patrasciou 4> who claims that the latter gauge is more suitable than the former. It is shown that there are two different types of constraints in the string model in accordance with Dirac's method.