Abstract The fatigue behaviour of plain (PC) and steel fibre reinforced concrete (SFRC) is usually assessed on the basis of experimentally derived SN curves. Test data exhibit large inherent scatter in numbers of cycles to failure leading to significant uncertainties in calculated lifetime predictions. The paper deals with the stochastic evaluation of these uncertainties introducing mechanically based SN curves. The curves take into account effects of the concrete's fracture energy and of the steel fibre's post cracking tensile strength as a function of fibre type, dosage, orientation and bond behaviour by a dimensionless ductility index. Response surface methodology serves to fit the approach to experimental data taken from the literature and to assess its prediction accuracy by specifying confidence intervals. Besides, variance‐based sensitivity analyses using Latin Hypercube Sampling to generate random variables are numerically executed to theoretically quantify the effect of scattering input parameters on scattering numbers of cycles to failure. Two examples demonstrate the procedure and underline the experimental observation that the concrete's tensile strength governs fatigue performance at high stress levels while fibre dosage and orientation become meaningful at low ones.