Currently, ASTM E739-10, Standard Practice for Statistical Analysis of Linear or Linearized Stress-Life (S-N) and Strain-Life (ε-N) Fatigue Data, provides analysis methods to determine confidence limits on the mean of linearized fatigue life data. Although these techniques are sufficient for the casual user, rigorous data analysis for structural applications typically involves the calculation of tolerance bounds for nonlinear data sets to ensure design criteria. The effect of nonlinearity becomes more significant as life predictions become significantly longer approaching the material fatigue limit of certain combinations of materials and test conditions (maximum stress, R-ratio, K_t, etc.). While analyzing numerous fatigue data sets, several issues have been seen that do not currently fit within the limitations associated with ASTM E739-10. These limitations include: (1) The treatment of data sets containing nonuniform variance over a range of stress levels needs to be considered. (2) Nonsymmetric distributions in non-transformed space (e.g., two-parameter Weibull or log-normal) may describe the failure distributions better for given stress levels and materials than a linear failure model in non-transformed space. (3) Data taken over a whole range of stress levels may provide unrealistic tolerance bounds for low stresses when compared to the development of tolerance bounds on a stress-level by stress-level basis. This study addresses some of these limitations using numerical simulation-based statistical distributions obtained from several relatively large data sets. These numerical simulations are analyzed using the linear methods shown in ASTM E739-10 as well as other techniques such as those outlined in the Metallic Materials Properties Development and Standardization (MMPDS) Handbook to develop confidence and tolerance limits.