Neuro-fuzzy learning algorithms with Gaussian-type membership functions based on the gradient descent method are well-known methods for generating fuzzy rules. In the conventional method, however, increasing the number of inputs greatly increases the number of parameters. Representation of fuzzy rule tables is thus difficult. We propose a new learning approach, the complex-valued neuro-fuzzy learning algorithm, which extends the conventional method domain to the complex numbers. In this method, inputs, antecedent membership functions, and consequent singletons are complex, and outputs are real. For parameter tuning, we use complex back propagation. The method assigns a two-dimensional real number to the real and imaginary parts of the complex number, which is used as a single complex-valued input. This process greatly reduces the number of tuned parameters, leading to the same or better learning than the conventional method. We compare the proposed and conventional methods using several function identification problems and show that the proposed method outperforms its counterpart, making it a useful tool for learning a fuzzy system model.