The authors aim to build a theoretical system for calibrating the position detection of a single quadrant photodiode (QPD) attached to 2D rotators and to suppress noise using a Sugeno‐type fuzzy inference system. According to the results, if the original ANFIS network (NWo) is trained under the conditions of resolution (rso), traing samples (No), and the possibility of random noise (po) and presented as {original, rso, No, po}, with the #X testing case presented as {case #X, rsx, Nx, px}, case #X is expected to be the all‐pass case of the two root‐mean‐squared error criteria under the conditions of rsx ≥ rso and px ≤ po. Thus, the measurement accuracy is improved without tracking historical data. Furthermore, the two conditions indicate the limitations of the proposed ANFIS‐based method. These conditions can be employed to save time and money. If a company produces many types of QPD rotators with different values of rs, the engineers can optimise the design to train the ANFIS network with only one case with the smallest rs and a larger p for all types of QPD rotators with any value of rs and p. This demonstrates the potential industrial application of the proposed method.