Abstract: An advanced description method of initial fatigue quality was proposed, in which neural network interpolation is employed instead of Lagrange interpolation to compute the values of the time to crack initiation (TTCI) and maximum likelihood estimation is used to estimate TTCI distribution parameters instead of mean rank estimation. Taking account into the randomicity encountered when only one group of TTCI values is used to estimate distribution parameters, several groups of TTCI distribution parameters were gained after maximum likelihood estimating for several groups of TTCI values corresponding to the given reference crack sizes. Then, equivalent initial flaw size distribution can be confirmed based on the relations of several groups of TTCI distribution parameters. The fatigue test of some component and its initial fatigue quality analysis show that the advanced method is feasible and reasonable.