The behavior of small cracks (less than 1 mm in length) have been shown to be quite different than large cracks for a variety of materials. In the past two decades, the large-crack test procedure (load shedding) has been shown to cause a load-history effect in the low-rate regime, generating elevated thresholds, and slower rates than steady-state behavior, which caused a large part of these differences. The literature has shown that small-crack data is more appropriate for damage tolerance and fatigue analyses. The objective of this work was to validate the development of artificial neural network (ANN) methods in fatigue crack growth predictions of small cracks. Two ANNs were developed: extreme learning machine (ELM) and radial basis function network (RBFN) to predict fatigue crack growth of small cracks for various materials. A wide range in stress ratio R and stress levels were considered for selected materials. The two ANNs were compared with each other in terms of mean squared error achieved and performance. The ELM method showed a superior interpolation and extrapolation ability compared to the RBFN method.