Predicting the quality of injection molded parts based on process data has been dealt with in research for some time. However, existing approaches did not prevail in the industry to date due to lacking integrity. An important step in building a successful quality model using machine learning methods is the selection of suitable process features. This paper describes the various methods and steps for feature selection, including filters, wrappers and embedded methods, and compares the resulting feature sets with regard to the achievable goodness of the quality models that were created using six different supervised machine learning algorithms for regression.