A very different suggestion to the Zadeh's rule for fuzzy inference is the least squares method proposed by T. D. Pham and Valliappan. The aim of this paper is to study the possible contradictions appearing in this method and, additionally, to extend it when functions are not necessarily monotonic.