Field inspection to detect rail flaws is performed primarily using ultrasonic transducers mounted on track-mounted vehicles (see Figure 1). Failure to promptly detect and repair these flaws imposes service reliability problems for both freight and passenger operations, and can pose safety concerns. Current ultrasonic rail flaw detection technology limits inspection vehicle operating speeds to about 15 mph and detects only about 70 percent of the flaws. Both of these restrictions result from the limitations of the human operator’s ability to interpret the signal from the detection equipment. When flaws are indicated with the current system, it is often necessary to stop the inspection vehicle and perform a hand test. This limits efficient management of rail repair and delays traffic. Neural networks have the potential to substantially enrich the capability to efficiently and reliably extract comprehensive information from the ultrasonic signal used to inspect rail, and to do so in a more automated manner.