The dispersion of the tidal volume and of the breathing frequency have been used to diagnose dysfunctional breathing during cardio-pulmonary exercise testing. No validated methods to objectively describe this dispersion exist. We aimed to validate such a method. We used simulations based on real-life parameters. Moving standard deviation (MSD) and residuals from locally estimated scatterplot smoothing (LOESS) were evaluated. The precision and the bias of each tested method at rest and during exercise simulations, with and without sighs, were measured. For LOESS, a 2nd degree polynomial was used, and different spans were tested (LOESS₁, LOESS_0.75, and LOESS_0.5). For MSD, different number of points used for the calculation were tested (MSD₇, MSD₁₁, MSD₁₅, and MSD19). The LOESS method was globally more precise, had less bias, and was less influenced by the trend as compared to MSD in almost all simulations except for extremely low dispersion combined with extreme trends. LOESS_0.75 had intermediate bias and precision between LOESS_0.5 and LOESS₁ in all simulations. LOESS_0.75 is a method that combines high precision, low bias, and low influenceability of trends. It could be considered as the method of choice to evaluate the dispersion of breathing parameters during cardiopulmonary exercise testing.