Abstract Fatigue investigation on 7050-T7651 chromic acid anodized aluminium alloy has been realized in order to develop a fatigue life predictive model which includes the different experimental aspects of fatigue cracking. In particular, it has been observed that, for this alloy and the sub-mentioned anodizing process conditions, degreasing process used after machining had no influence on fatigue resistance; the essential of the important fatigue resistance decrease was due to pickling process, leading to production of numerous pits which were found to be crack initiation sites; then anodizing process was responsible for a supplementary increase of fatigue resistance decrease. Failure surfaces observations have shown that these pits were responsible for cracks and numerous crack coalescence phenomena have been pointed out. The fatigue life predictive model presented in this paper is based on Suraratchaï’s model, developed at ICA-Toulouse, which is built on the stress concentration effect of machining surface roughness: here pickling pits are considered as stress concentrators from which cracks could occur if stress intensity factor is high enough; considering the very low size of pickling pits, half size of middle recrystallized grain size, this implies the introduction in the model of short cracks considerations such as short crack stress intensity factor range threshold and short crack growth rate law. Coalescence conditions are also introduced. Calculation are conducted from a 3D finite elements built from experimental topographic measurements which allow the determination of stress concentration coefficients induced by machining surface roughness and the presence of the pits. Anodized surface are rebuilt from pickled surface in order to overturn the lack of accuracy during topographic surface measurements. Predictive results fit very well with experimental results whatever the surface state.