This paper proposes an original method for characterising the Low Cycle Fatigue (LCF) lifetime using probability density functions. The protocol is based on statistics of microstructure heterogeneities taken as damage initiation sites, a qualitative mechanical analysis of the heterogeneities harmfulness and the definition of a micro-crack growth law. The technique is here established and the associated model identified for a nodular cast iron where the graphite nodules are assumed to be the damage initiation zones. The LCF lifetime is characterised from both a large set of experimental test between 300 and 600 • C and damage observations at the micro-scale. Experimental post-mortem observation combined with a simple numerical study first enable to assume the harmfulness of nodules according to their size and their probable role in the damage process A probability density function for the lifetime is then built from the following steps: (i) a quantitative analysis of the material micro-structure, which provides the probability density of nodules occurrence depending of their size (ii) an extreme value analysis using a Gumbel distribution and (iii) a micro-crack growth law associated with LCF conventional terms of energy densities. Its parameters are obtained using an optimisation process applied to laboratory fatigue experiment. The obtained probability function provides a good match for the lifetime and greatly improves results given by conventional criteria. It moreover provides a robust estimate of the lifetime scatter for different types of fatigue tests.