Time-series data is widely adopted in condition monitoring and fault diagnosis of wind turbines as well as other energy systems, where long-term dependency is essential to form the classifiable features. To address the issues that the traditional approaches either rely on expert knowledge and handcrafted features or do not fully model long-term dependencies hidden in time-domain signals, this work presents a novel fault diagnosis framework based on an end-to-end Long Short-term Memory (LSTM) model, to learn features directly from multivariate time-series data and capture long-term dependencies through recurrent behaviour and gates mechanism of LSTM. Experimental results on two wind turbine datasets show that our method is able to do fault classification effectively from raw time-series signals collected by single or multiple sensors and outperforms state-of-art approaches. Furthermore, the robustness of the proposed framework is validated through the experiments on small dataset with limited data. •An end-to-end Long Short-term Memory (LSTM) model for fault diagnosis of wind turbine.•Features directly learnt from multivariate time-series with no need of handcrafted ones.•LSTM captures long-term dependencies through recurrent behaviour and gates mechanism.•The proposed method outperforms state-of-the-art methods on two wind turbine datasets.•The robustness is validated through experiments on small dataset with limited data.