In this study, seven types of load controlled fatigue tests were carried out under variable stress amplitude in order to clarify the stress-strain response and the fatigue strength of SS 41 and STPY 41 plate specimens. The tests were conducted in constant stress amplitude loading, step stress amplitude loading, repeated high-to-low and low-to-high block loading, two types of block-random loading and random loading with exponential distribution. In each test, stress-strain loops were recorded with X-Y plotter from the start of cycling to visual crack initiation by using the strain gauges put them on the test piece. As a result, the following conclusions were obtained. (1) Sa-εpa (stress amplitude-plastic strain amplitude) curves are different with each variable loading patterns. (2) Cyclic softening for random loading is greater than that for constant stress amplitude in the conditions of Sa≥σy and Sa≤Saw, while the relation is contrary to above in the conditions of Sawa<σy, where Saw is a fatigue limit and σy is a static yield stress of the material. (3) Sa-εpa curve for random loading with exponential distribution can be obtained approximately by the step stress amplitude loading. (4) Fatigue life estimations by Miner's rule and Modified Miner's rule make over estimation compared with experimental results for block, block-random and random loadings. (5) Fatigue life estimations by Manson-Coffin's law do not coincide with the experimental results for block-random loading and random loading with exponential distribution included the large repeated number of stress amplitudes below fatigue limit. (6) Fatigue life estimations by Corten-Dolan's method fairly coincide with experimental results for various kind of variable stress amplitude loading.