King County, Wash., detective Robert Keppel was hunting a serial murderer in 1974. Working from the leanest of clues, his staff assembled some 30 lists of potential suspects. One named 4,000 University of Washington classmates of a female victim. Another, every patient released from the state’s mental wards in the preceding decade. Using a punch-card computer, the investigators compared the lists. Among the 25 that turned up most often was one now-infamous name: Ted Bundy.

Keppel, currently chief criminal investigator with the Washington State attorney general’s office, says that rudimentary program was the first ever written to catch a killer. Today, Keppel is again helping to pioneer crime-fighting computing by testing a new system developed at the Pacific Northwest National Laboratory in Richland, Wash. Called Computer Aided Tracking and Characterization of Homicides (CATCH), it uses a neural network to discover unseen patterns in the state’s computerized murder-investigation records.

Finding patterns in complex data is work that humans excel at-but computers don’t. Neural networks, however, process complex information much as the human brain does, says CATCH project leader Lars Kangas.

With funding from the National Institute of Justice, Kangas’ team fed data on 5,500 Washington homicides-solved and unsolved-into CATCH. The program clustered similar crimes based on over 200 variables. Homicides grouped together could be the work of one person. CATCH can also generate a suspect profile by comparing unsolved cases with similar solved cases.

Keppel’s crime analysts are studying the CATCH matrices to see if the neural network’s hunches were right. So far, the results are mixed: two murders committed by the same person ended up on opposite sides of the matrix, but elsewhere CATCH correctly grouped known serial crimes. Though it’s too soon to know if CATCH will make a decent detective, Keppel remains hopeful that the program will help finger monsters like Ted Bundy lurking in the data.