For a given problem, a condition number is a quantity depending on the data that measure the numerical sensitivity of the data to perturbations. This parameter plays a fundamental role in the complexity analysis of numerical algorithms, both from a run-time and precision control perspective. However, because of this, numerical algorithms tend to have complexity estimates that do not depend solely on the input size. The main philosophy to solve this is to perform a probabilistic analysis of the condition number assuming some reasonable probability distribution of the input. In this talk, we introduce the different ways in which such a probabilistic analysis can be done and the differences between the different approaches.