Recently, we have proposed a novel intuitionistic fuzzy inference system (IFIS) of Takagi-Sugeno type which is based on Atanassov’s intuitionistic fuzzy sets (IF-sets). The IFIS represent a generalization of fuzzy inference systems (FISs). In this paper, we examine the possibilities of the adaptation of this class of systems. Gradient descent method and other special optimization methods are employed to adapt the parameters of the IFIS in regression problems. The empirical comparison of the systems is provided on several well-known benchmark and real-world datasets. The results show that by adding non-membership functions, the average errors may be significantly decreased compared to FISs.