An efficient algorithm for condition monitoring of rotating machines is proposed in this paper. Condition indicators are derived from sound signals, and used to arrive at a decision about the performance state of the machine. Sound signals are recorded by microphones and processed using time-frequency domain analysis. In this study, number of statistical features; such as mean, standard deviation, skewness, and kurtosis are considered. These statistical features were proven to be effective and simple to interpret. Healthy, about to be faulty, and faulty performance states of the machine are considered, and audio signals are recorded for each state. The five main steps comprising the implemented approach are data acquisition, preprocessing, feature extraction, time and frequency domain analysis, and the decision making. Based on the adopted statistical measures, the experimental results indicate that an excellent recognition of machine performance states is obtained, leading to an efficient fault detection and diagnosis.