Six teams tested recommended procedures for estimating fish kill magnitude during a simulated fish kill in a small east Texas stream. Dead fish were added to the stream and counted 16 h later using accepted guidelines for sampling and counting dead fish. Thirty-one percent of the 943 dead fish that were initially placed in the stream remained when teams counted fish. The highest estimate of dead fish included only 39% of the total number of dead fish added to the stream. Independent estimates were moderately variable (mean = 308, SD = 54) and accurate when compared with the number of carcasses remaining in the stream (N = 291) at the time of the investigation. Small individuals (<15 cm total length, TL) and less abundant species were underestimated to a greater degree than large individuals (>35 cm TL) and more abundant species. Counts involving fewer fish resulted in a greater underestimation of fish killed than did counts of larger numbers of fish. Complete enumeration gave the most accurate estimate of the number, species, and size distribution of carcasses remaining in the stream. However, all estimates grossly underestimated the total number of dead fish placed in the stream. Scavengers probably removed most of the dead fish during the 16 h before the investigation.