This thesis deals with the analysis of the influence of noise on the robustness of selected global optimization methods. In practical engineering and scientific problems, objective function values are not always exact, because they often come from measurements, numerical simulations or stochastic models. This issue is especially important in black-box optimization, where the optimizer works only with evaluated objective function values and does not have access to exact derivatives. The aim of the thesis is to compare the behaviour of selected optimization algorithms under noisy objective function evaluations. The experimental part was carried out using the COCO benchmarking platform with the standard BBOB test suite, and the results were processed using the COCOPP tool. Noise was introduced into the problems using the cocoex.noiser.Noisifier module, which makes it possible to test algorithms under controlled conditions without changing their interface. The experiments were implemented in Python and used optimizers mainly from the Nevergrad library. In total, 23 algorithms were compared under additive and subtractive Cauchy-type noise. The probability of noise occurrence was gradually varied from the noise-free case to fully noisy evaluation. The results showed that algorithm robustness strongly depends on the type of noise, its probability and the dimension of the solved problem. In the noisefree case, the best results were achieved mainly by CMA-based methods. At higher levels of additive noise, Differential Evolution variants were competitive in lower dimensions, while Powell-based methods proved to be more robust in higher dimensions. Subtractive noise generally had a stronger negative effect already at lower probabilities. The main contribution of the thesis is the creation of a reproducible experimental procedure for evaluating the robustness of optimization algorithms in the COCO environment. The obtained results confirm that none of the tested methods is universally best, and that the suitability of an algorithm depends on the character of the problem as well as on the type of noise.