A probabilistic fatigue crack growth (FCG) life prediction is conducted on a turbine disc considering small sample data involving material properties and geometric dimensions in this study. Wiener process with two transformed time scales is used to quantify the time-varying dispersion of FCG degradation process. The accuracy of Weiner process model is verified by using the FCG experimental data for standard compact tension (CT) specimens. Then, FCG tests on the component-like specimens are conducted to establish the time-varying FCG model considering geometrical features. During this process, Bootstrap method is introduced to expand the small test samples into larger Bootstrap samples. Meanwhile, the dispersion for geometric dimensions of the small sample is described by using probability box method. Finally, probabilistic FCG analysis is conducted on a turbine disc through using quasi-Monte Carlo sampling method. Results show that the average relative error between the lower bound of Ns and Na is within 20%, and the average relative error between the upper bound of Ns and Na is less than 15%, which further demonstrates the accuracy of probabilistic FCG method.