The success of a product line is the ability to improve application engineering, heavily depends on the quality of Product Line Models (PLMs). This paper reports on our effort to develop a method that exploits mining techniques such as the apriori algorithm, independence tests and the like to automate the construction of a PLM specified with FORE, starting from a collection of Product Models (PMs). Using these techniques, the proposed method guides the identification of candidate features, group cardinalities and dependencies. These can be used to progressively construct the PLM consistently with the existing PMs. The method was developed and tested in an industry setting starting with bills of materials as a collection of PMs. One interesting lesson learn from this experiment is that while the PLM is constructed, the domain engineer discovers errors in PMs. We believe that this advocates for a tighter intertwining between domain engineering and application engineering.