Over the years, numerous researchers have attempted to develop a method for detecting the interface between coal, roof, and floor during the mining operation. The complex nature of the geological interface has proven to be a formidable problem for any type of coal interface detection system. A novel approach to detecting coal-roof and coal-floor interfaces during mining operations, currently being pursued by the Bureau of Mines, is based on a seismic acquisition method integrated with various signal-processing techniques. Seismic transducers (accelerometers) are affixed to the roof, coal, and floor at a distance from the mining machine. Unique seismic signals are generated during mining, depending upon which stratum is being cut. A discriminator extracts and stores mathematical parameters characteristic of the signals in the form of an adaptive learning network (aln). The system is initially "trained" to recognize three cases by intentionally cutting some sample roof, coal, and floor. The computer uses this aln to tell where new unknown signals associated with the mining process are originating. Via a feedback loop, the computer has potential for keeping the machine in the coal seam. Described are the in-seam seismic concept, the experimental setup, the data analysis methods being used, and some preliminary findings.