The purpose of this paper was to decompose experimentally the variability of the fatigue crack propagation rate da/dN into two parts, the inter-specimen variability and the intra-specimen variability. Specifically, in order to separate σb2 and σw2 experimentally, fatigue crack propagation tests were conducted under the condition of constant ΔK, where σb2 and σw2 were the inter-specimen variance and the intra-specimen variance of logda/dN, respectively. The condition of constant ΔK was attained by use of the tapered double cantilever beam specimens. The material tested was the high strength steel HT60. By measuring the increment of the crack length Δa at regular cycle intervals of ΔN, da/dN was evaluated as da/dN=Δa/ΔN as in usual cases.ΔK was chosen as ΔN=5000 cycles. The technique of analysis of variance was applied to the test data to estimate σb2 and σw2. It was shown that σb2 and σw2 were of the same order. The F test also showed that the inter-specimen variability could not be neglected.