Previous studies evaluating computer algorithms for endocrine pulse detection have estimated the rate of false-positive pulses in series of purely random variations (i.e., “noise”) and have determined pulse-detection criteria associated with low levels of such false-positive rates. The present study investigates the relationship between the false-positive rate and the sizes of the false-positive and false-negative errors on pulse frequency for series including both pulses and noise. The algorithm used (ULTRA) proceeds by eliminating all peaks of concentration for which either the increment or the decrement does not exceed a threshold expressed in multiples of the local intra-assay coefficient of variation. A total of 336 computer-generated series was analyzed using thresholds of two and three coefficients of variation. The effects of noise level, pulse frequency, pulse amplitude, and presence of a base-line variation on the sizes of the false-positive and false-negative errors were evaluated. The false-positive rate in noise series exceeded the false-positive rate by a 4- to 10-fold factor in series including at least 8 pulses/100 samples. When pulse frequency increased, the false-positive error decreased, but the false-negative error increased. In series with more than 8 pulses/100 samples, the use of thresholds aimed at maintaining the false-positive rate in noise series below 1% resulted in a false-negative error in excess of 20%. In conclusion, for hormonal profiles that include 8 or more pulses/100 samples, the use of pulse-detection criteria tailored to minimize the false-positive rate in noise series may result in an underestimation of pulse frequency.