Assume a scenario where among the chief complaint strings of two unrelated patients in the same District on the same date there was a mention of bloody stools in pediatric cases. The multiple mentions of “bloody stools” or “pediatric” might not be surprising, but the tying together of these two factors, given matching geographic locations and timings of reporting, is sufficiently rare that seeing only two such cases is of interest. This was precisely the evidence that was the first noticeable signal of the tragic Walkerton, Canada, waterborne bacterial gastroenteritis outbreak caused by contamination of tap water in May 2000. That weak signal was spotted by an astute physician, not by a surveillance system. Reliable automated detection of such signals in multivariate data requires new analytic approaches.