Modelling fatigue crack propagation by damage accumulation at the crack tip originally proposed by McClintock is reworked out using metallurgical considerations. On a physical basis it is shown that the validity of such models is actually confined to the low crack growth rate range corresponding to stage I growth such as along crystallographic planes. At higher crack growth rates there is a transition to plastic stretching mechanisms which could be described by crack opening displacement models. An evaluation of two models has been carried out on a 33Co Ni alloy where extensive information was available: both are based on the strain singularity as computed by Tracey from a finite element analysis of plane strain small scale yielding and as adapted to cyclic loading under Rice's hypothesis and taking the grain as the critical element below which size continuum mechanics do no longer apply. It is pointed out that the use of this strain singularity which is not able to account for crack closure, renders models unable to predict experimental crack growth rates curves but only intrinsic curves relating the growth rate to the effective stress intensity amplitude as defined by Elber. The first model which assumes fatal cracking of a grain with an average uniform cyclic equivalent strain, underestimates the crack growth rates and in addition it yields results very sensitive to the shape and size of the grain. The second model which assumes progressive cracking in a grain with a cyclic equivalent strain gradient is able to predict the intrinsic crack growth rate curve in the 33Co Ni alloy and to yield predictions consistent with other experimental data.