Most of actual service loads for real structures are random in general and, moreover, are stationary processes in many cases. It is essential for the structural engineer to be able to estimate fatigue strength of structural members under the effect of such randomness in service loading on the basis of the conventional fatigue test data. Engineers are sometimes confronted with the necessity of prompt relevant action on cracks discovered by inspection in service or in fabrication. A number of investigations on estimating methods for fatigue crack growth under random loading have been carried out, but most of them are based on the assumption that the random loading can appropriately be replaced by so-called “block loading”. Experimental investigations on the comparison between fatigue behaviors under purely random loading and block loading and/or constant amplitude loading are quite limited. In this paper, the authors have conducted fatigue crack propagation tests using a computer aided testing system, where random process required was generated by micro-computer, as well as block loading and constant amplitude tests. Experimentally obtained features and comparative discussion based on the experimental data are given.