The length-scale dependence of fatigue crack growth is evaluated for a set of metallic materials, namely titanium Ti-6Al-4V, ductile iron EN-GJS-500-7 and tool steel AISI H13, by performing fatigue crack growth tests on geometrically similar compact C(T) specimens of different sizes. With references to length-scale-invariant variables, notably the crack growth rate normalised by the specimen width, it is demonstrated that fatigue crack growth is not a length-scale-invariant process for the tested conditions. In particular, the length-scale dependence is less significant for larger specimens and at longer crack lengths. The test results also contradict the hypothesis of similitude, i.e., that the growth rate is uniquely related to the stress-intensity-factor range, as smaller specimens manifest a higher growth rate when compared at the same stress-intensity-factor range. The observations are in line with presented fracture-mechanical demonstrations, which show that the Paris–Erdogan law depends on the length scale whenever fatigue crack growth is anticipated to be scale invariant.