In a real word optimization problem, it is common that the only available data are given by an oracle, which receives as an input a feasible point and outputs the value of the objective function at that point. In particular, no extra knowledge about the internal process involved in the objective function – such as the analytical form, some gradient or Hessian – can be exploited. This is termed black-box optimization. Additionally, the returned fitness value might be perturbed or inaccurate, due to some measurement errors, to the sensor’s sensitivity or to other stochastic effects: this is termed noise. We will give a few introductory examples of noisy black-box problems then present different families of noisy black-box optimization algorithms: stochastic vs. deterministic algorithms, value-based vs. comparison based algorithms. We will discuss their performances, depending on the objective function characteristics and the noise model, but also depending on the evaluation criteria that is taken into account.