The fast growing wind industry requires a more sophisticated fault detection approach in pitch-regulated wind turbine generators (WTG), particularly in the pitch system that has led to the highest failure frequency and downtime. Improved analysis of data from Supervisory Control and Data Acquisition (SCADA) systems can be used to generate alarms and signals that could provide earlier indication of WTG faults and allow operators to more effectively plan Operation and Maintenance (O&M) strategies prior to WTG failures. Several data-mining approaches, e.g. Artificial Neural Network (ANN), and Normal Behaviour Models (NBM) have been used for that purpose. However, practical applications are limited because of the SCADA data complexity and the lack of accuracy due to the use of SCADA data averaged over a period of 10 min for ANN training. This paper aims to propose a new pitch fault detection procedure using performance curve (PC) based NBMs. An advantage of the proposed approach is that the system consisting of NBMs and criteria, can be developed using technical specifications of studied WTGs. A second advantage is that training data is unnecessary prior to application of the system. In order to construct the proposed system, details of WTG operational states and PCs are studied. Power-generator speed (P-N) and pitch angle-generator speed (PA-N) curves are selected to set up NBMs due to the better fit between the measured data and theoretical PCs. Six case studies have been carried out to show the prognosis of WTG fault and to demonstrate the feasibility of the proposed method. The results illustrate that polluted slip rings and the pitch controller malfunctions could be detected by the proposed method 20 h and 13 h earlier than by the Al approaches investigated and the existing alarm system. In addition, the proposed approach is able to explain and visualize abnormal behaviour of WTGs during the fault conditions. (C) 2017 Elsevier Ltd. All rights reserved.