This thesis has two parts. In the first part, we use the results of new fatigue experiments conducted with variable load levels as well as variable stress ratios to critically assess three (two old and one relatively new) cumulative fatigue damage models. These models are deterministic. Such models are usually tested using multiple blocks of periodic loading with differing amplitudes. However, available data pertains to zero-mean loading, and does not investigate the role of variable stress ratio (Smin/Smax). Here, we present experimental results for variable stress ratios. Two specimen geometries and two materials (Al 2014and Al 2024)are tested. Manson’s double linear damage rule (DLDR)gives the highest accuracy in predicting the experimental outcome, even in the presence of variable stress ratios, whereas predictions of the newer model (“A constructive empirical theory for metal fatigue under block cyclic loading,” Proceedings of the Royal Society A, 464 (2008), 1161-1179) are slightly inferior. The widely used Miner’s rule is least accurate in terms of prediction. The merits and drawbacks of these models, in light of the experimental results, are as follows. The DLDR, though accurate, has minor scientific inconsistencies and no clear generalization. The constructive model has possible generalizability and more appealing scientific consistency, but presently has poorer accuracy. Miner’s rule, least accurate, lies within the constructive approach for special parameter values. The DLDR can guide the new (constructive)approach through new parameter fitting criteria. In the second part of this thesis, we consider the scatter in fatigue life and use the Weibull distribution to describe ‘S-N-P’ curves. We first assume homoscedasticity (load-independent or constant variance) and present a way to draw a p-percentile line on a log-log load-life plot. Then heteroscedasticity (load-dependent variance) in fatigue life is incorporated and a simple statistical model is proposed, to obtain a straight line percentile plot at a pre-specified probability of survival ps. The proposed method is illustrated for Al 2014-T6 and Al 2024-T4 data sets (extracted manually) from MMPDS-01 (a data handbook).