In this paper we consider the class of stochastic stationary sources induce
d by one-dimensional Gibbs states, with Holder continuous potentials, we sh
ow that the time elapsed before the source repeats its first it symbols, wh
en suitably renormalized, converges in law either to a log-normal distribut
ion or to a finite mixture of exponential random variables. In the first ca
se we also prove a large deviation result.