Common applications of Condition-Based Maintenance utilize sensors mounted on mechanical components to diagnose failure conditions and incipient faults. The algorithms which process the raw data into diagnostic and prognostic indicators are typically derived from theoretical models, traditional signal processing metrics, or through trial-and-error observations. Condition monitoring devices are becoming increasingly common and are producing large volumes of data, yet relatively few studies have examined this real-world field dataset. Utilizing common data mining algorithms, an inferential study was performed to create diagnostic indicators which can distinguish healthy and faulted drive train components. Data was collected from multiple rotating components from several hundred rotorcraft and from a laboratory setup of the same components. Preconditioning filters for noise reduction and dimensionality reduction were performed, and it was found that for all components in the study, it was possible to represent complex vibration spectra with a relatively few number of attributes. Uniqueness among the individual articles within the sample population was also observed, and the implications of these findings are discussed. From the results of the preliminary investigation, classifier models were built on laboratory and field data to identify components which were normal, nearing failure, or failed. Multiple evaluations of the classifiers were performed, and a general approach to achieve data-mining derived condition monitoring is proposed.