The author discusses the resistance of metal to the propagation of cracks, and points out that in the accepted method of fatigue testing the results are not conclusive because there is no distinction between (1) the load and the number of repeated stresses required to start cracks and (2) the load and the number of reversals needed to propagate the cracks to failure. The investigation reported in the paper was undertaken to determine (a) the stress and number of reversals required to start a fatigue crack, and (b) the resistance of the material to the effect of a fatigue crack once it had been formed. The tests reported were conducted on annealed cold-rolled low-carbon steel bars with various finishes. The author presents data on the rate of progress of the cracks and on the size of the cracks, the latter of which was determined by magnetizing the bars and inspecting them with finely divided iron oxide in a bath of kerosene.