Condition monitoring is a major part of predictive maintenance which monitors a particular condition in machinery to identify changes that could indicate a developing fault. It allows maintenance to be scheduled and preventive actions to be taken to reduce the failures. This study presents a new feature extraction method that is used to detect the faults of worm gears (WG) during the condition monitoring process under various operating conditions. In this study, an experimental setup that can operate under different operating conditions has been developed to obtain vibration and acoustic data. The feature extraction technique Common Spatial Pattern (CSP) has been used for the first time to detect the faults (wear, pitting and tooth breakage) of machinery from vibration and acoustic data. Fault detection and classification were performed with Artificial Neural Network (ANN), Support Vector Machine (SVM) and K-Nearest Neighbour (k-NN) methods based on CSP features obtained using vibration and acoustic signals. According to the classification performance results, ANN method has produced considerable high accuracies for two class and multiclass classification when compared with the Support Vector Machine (SVM), K-Nearest Neighbour (k-NN). Moreover, the ANN classification results have also been compared with the Convolutional Neural Networks (CNNs) results in the literature. Finally, the performance of CSP features was validated with the commonly used time and frequency domain features. The contribution of this work includes the first time usage of CSP features for fault detection which were extracted from vibration and acoustic data of an experimental WG set. Moreover, various fault types of WGs under changing loading and speed have been examined for the first time. The results show that ANN with CSP features could achieve excellent performances in condition monitoring of WGs under variable operating conditions.