In manufacturing processes, surface finish of a product is very crucial in determining the quality. Therefore, the surface quality including the surface roughness is still the most important obstacles against the incremental sheet metal forming (ISMF) process. As a consequence, the possibility to predict the surface roughness values in incremental forming and to correlate these values with the forming parameters can be useful in order to control this important target. Accordingly, an adaptive neuro-fuzzy inference system (ANFIS) is used to predict the surface roughness of parts produced by single-point incremental forming (SPIF) process. The hybrid learning algorithm is applied in ANFIS to determine the most suitable membership functions (MFs) and to simultaneously find the optimal premise and consequent parameters by directly minimizing the root mean squared error (RMSE) as a performance criterion. In order to achieve this target, five forming parameters, namely (tool diameter, incremental step size, tool shape, rotational speed and slope angle) are studied to form pyramid like shapes for the purpose of roughness measurement. Experimental results show that the difference sigmoidal MF gives the minimum RMSE. The predicted surface roughness values using ANFIS are compared with actual data. The comparison indicates that the utilization of difference sigmoidal MF in ANFIS could achieve a satisfactory prediction accuracy using both training and testing data when this MF is adopted. The training and testing prediction accuracy are 95.972% and 85.799% respectively.