Cyclic tests with or without tensile holding periods were conducted in air at 823 K on a modified 9Cr–1Mo martensitic steel. In addition to stress–relaxation fatigue (RF) tests with a hold time at maximum load, creep–fatigue (CF) experiments were carried out. These CF tests were strain-controlled during the cyclic part of the stress–strain hysteresis loop and then load controlled when the stress was maintained at its maximum value to produce a prescribed value of the creep strain before cyclic deformation was returned under strain-controlled conditions. This unusual testing procedure enabled larger viscoplastic strains to be reached during the holding period than during usual relaxation–fatigue (RF) tests. The relationship between the number of cycles to failure of pure fatigue tests and the cyclic strain range is established for pure fatigue tests. The lifetime reduction due to holding periods is highlighted and quantified. The fatigue lifetime reduction due to holding periods is all the more pronounced as the cyclic strain amplitude is low. No creep cavitation is visible by microscopic observations, while the environment is found to play a key role in damage accumulation and interaction. Two main failure mechanisms are observed depending both on the fatigue strain range and on the duration of the holding period. An attempt is made to explain the existence of these two domains in relation with oxidation effect.