Knowledge acquisition is one of the most important issues in the development of intelligent systems. A good understanding of the investigated domain often proves crucial for systems that deal with large datasets of structurally complex objects, e.g. Optical Character Recognition (OCR) systems. The central issue in such systems is the construction of classifiers within vast and poorly understood search spaces, which is a very difficult task. Nonetheless this process can be greatly enhanced with knowledge about the investigated objects provided by a human expert. We propose a framework for the transfer of such knowledge from the expert and show how to incorporate it into the learning process of a recognition system using methods based on rough mereology. We also demonstrate how this knowledge acquisition can be conducted in an interactive manner, with a large dataset of handwritten digits as an example.