We present an approach to the cell formation problem, known from group technology, which is inspired by formal concept analysis. The cell formation problem consists in allocating parts (objects) to machines (attributes), based on the machine-part matrix. This can be viewed as forming groups consisting of a set of parts and a set of machines. Such groups resemble formal concepts in the input data. Due to the specific nature of the performance assessment in the cell formation problem, good groups can be thought of as rectangles which, unlike those corresponding to formal concepts, contain a few blanks, i.e. which are not full of crosses in terms of formal concept analysis. Moreover, such groups need to be disjoint both in terms of objects and attributes. In this paper, we present an algorithm for the cell formation problem, experimental results, and a comparison to some methods proposed in the literature.