<p>Fatigue reliability analysis based upon limited and uncertain data brings uncertainties in the inputs such as probability distributions and their respective parameters. However, in practice, the data is compiled by conducting physical tests. The uncertainties based on limited physical test data need to be carefully evaluated. Underestimation or overestimation of reliability based upon uncertain data and the variability in experimental conditions needs to be evaluated. The uncertainties pertaining to distribution are mitigated using statistical tools. Literatures are available wherein either parametric or non-parametric distributions are used to estimate reliability for uncertain data. However, this paper attempts to use both parametric as well as non-parametric distributions on a set of uncertain data and tries to compare the reliability. First, the experimental data is assumed to follow the Weibull and Lognormal distribution. The fit of these distributions with the assumed distributions are evaluated and then the reliability is estimated for these distributions. Since the data is limited and uncertain, a non-parametric estimator such as the Kaplan–Meier estimate is used to compute reliability. The approach is applied on steel plate welded joints and the data on a number of cycles up to failure was studied. This study shows that when dealing with limited and uncertain fatigue data, the choice of failure distribution significantly affects the reliability estimate. Comparing parametric (Weibull and Lognormal) and non-parametric (Kaplan–Meier) methods indicates that each captures different aspects of data uncertainty. Using both approaches provides a more reliable and balanced interpretation of fatigue behavior than relying on any single model.</p>