A major asset of fuzzy logic systems is dealing with uncertainties arising in their various applications, thus it is important to make them achieve this task as effectively and comprehensively as possible. While singleton fuzzy logic systems provide some capacity to deal with such uncertainty aspects, non-singleton fuzzy logic systems (NSFLSs) have further enhanced this capacity, particularly in handling input uncertainties. This paper proposes a novel approach to NSFLSs, which further develops this potential by changing the method of handling input fuzzy sets within the inference engine. While the standard approach is getting the maximum of the intersection between input's and antecedent's fuzzy sets (in the “pre-filtering” stage), it is proposed to employ the centroid of the intersection as the basis of each rule's firing degree. The motivation is to capture the interaction of input and antecedent fuzzy sets with high fidelity, thus making NSFLSs more sensitive to the input's uncertainty information. The testbed is the common problem of Mackey-Glass time series prediction in the presence of input noise. Analyses of the results show that the new method outperforms the standard approach (by reducing the prediction error) and has potential for a more efficient uncertainty handling in NSFLS applications.