The uneven distribution of flow phases in evaporator channels can drop the heat exchanger efficiency up to 30%. Due to its dependence on the interaction of several coexisting variables – both geometry, operating conditions, and fluid properties – it is a complex phenomenon to analyze. Most studies focus on the effect of single parameters: this is an inefficient and expensive way of doing experiments, and the results lack in understanding how the combination of variables affects the flow distribution. This paper presents a methodology to optimally characterize and predict the distribution of flow phases in the channels of an evaporator header based on Design of Experiment (DoE) techniques. Despite the proven potential of DoE methods, they have never been applied in this field. Tests were conducted with an air–water mixture in the configuration horizontal header with vertical channels with downward flow, varying inlet pipe position, channels intrusion, presence of a splashing grid at the header inlet, and air and water flow rates. Results prove that, when working with complex processes, interaction effects between variables cannot be neglected as they significantly affect the response. The most affecting parameter was found to be the air flow rate, followed by the combination between inlet pipe position and presence of the splashing grid. With horizontal inlet, the optimal response was given by absence of intrusion, presence of the splashing grid, lowest water, and highest air flow rate. Instead, for the vertical case, the distribution was enhanced with the highest intrusion, absence of the grid, and highest water and air flow rates. Lastly a first attempt to model the process was performed. Even if a universal regression model has low accuracy (51%), restricting the area of analysis can result in valid predictive relations, with accuracies up to 91.4%.@en