A statistical process, and system for implementing the process, is described for the analysis of stress waves generated in operating machinery or equipment. This technique is called Probabilistic Stress Wave Analysis. The process is applied to a population of individual "feature" values extracted from a digitized time waveform (such as a 2 second Stress Wave Pulse Train, or a 2 month history of Stress Wave Energy). Certain numeric descriptors of the statistical distributions of computed features are then employed as inputs to decision making routines (such as neural networks or simple threshold testing) to accurately classify the condition represented by the original time waveform data, and thereby determine a status of the operating machine/equipment.