A fatigue model for unidirectionally reinforced composites that are loaded parallel to the fibers is proposed that is based upon the growth of interfacial fatigue cracks. Certain predictions of the model are verified by experimental results, namely, that the rate of crack growth is independent of crack length and that the fatigue life at >10⁶ cycles is relatively insensitive to the mean stress. Depending upon whether the matrix is elastic or elastic-plastic, a different dependency of the rate of crack growth on stress results, but the data available are insufficient to make a precise distinction between the two cases. In addition, at higher stress amplitudes the tensile rupture mode of separation becomes increasingly important, a fact dealt with only empirically by the model. It is concluded that for loading parallel to the fibers the most fatigue resistant composites will consist of strong, uniform, and defect-free fibers in a relatively soft matrix.