The purpose of this work is to propose a method of global characterization of the behaviour of pressure-swing-adsorption (PSA) cycles based on statistical experiments design, the ＇experiments＇ being actually numerical (simulation). The result of that procedure presented here is a set of polynomials describing the performances of the PSA cycles (purity, yield, productivity) as functions of operating or design variables (flow rates, pressures, composition, bed length, step durations). These polynomials represent the sensitivity of the position, bed length, step durations). These polynomials represent the sensitivity of the process to fluctuations, perturbations or international changes of process variables, and can be used, within their range of validity, as a fast aid to operators, and as a tool for design and optimization. The specific case investigated is a PSA cycle for pure hydrogen production from a mixture with CO2, CH4, CO and N2 obtained by cracking natural gas. The process comprises two adsorbent beds, each containing a zeolite layer on top of an active carbon layer.