Thermal fatigue crack propagation under random temperature fluctuation is theoretically investigated from a probabilistic view point by the use of a Markov approximation method, under the condition that the temporary variation of the inner surface temperature of plate is modeled as a wide-band stationary Gaussian process. First, a crack growth equation is formulated on the basis of the Paris law under the assumption that the stress intensity factor range ΔK can be approximated by the local expectation of a relative maximum of the stress intensity factor K. Next it is extended to a random differential equation, where the randomness in crack propagation resistance is taken into account. The Markov approximation method is then applied to derive a residual life distribution function as well as a probability distribution function of the crack length. Finally, numerical examples are shown to examine the quantitative behavior of the residual life distribution, whose results indicate that the present model is applicable even if the spectrum of temperature is of narrowband type.