Wind turbines operate almost uninterruptedly, and their operation is often subject to harsh environments, as well as complex and dynamic loads. Fourier analysis, a standard diagnostic technique, presents some limitations regarding the use of non-stationary, non-periodic, noisy data, which is precisely the case with wind turbine data. Due to these limitations, unseen faults could progress and cause severe, and even catastrophic, failure in wind turbines. Information theory quantifiers, such as entropy, divergence, and, statistical complexity measure, are proposed to evaluate the health status of wind turbine components. In this work, this is done via the decomposition of the signal in time, frequency, and timefrequency domain, namely via Bandt and Pompe, power spectrum, and wavelet packet decomposition. Two different real data sets from operational wind turbines were characterized by the proposed methods. Results demonstrate that the proposed method can distinguish (cluster) well between the states of fault, but also presented some limitations, mainly related to the complexity of the signal from operational wind turbines. Based on these results, new methods, complementary to Fourier analysis, are proposed to be employed in wind turbine data, aiming to increase the capability of detecting faults in such a complex environment. (C) 2020 Elsevier Ltd. All rights reserved.