A versatile method is presented for generating synthetic hormonal time series, containing peaks at known locations, to be used to objectively evaluate both the false-negative (F-) and false-positive (F+) statistical error rates of computerized pulse-detection algorithms. Synthetic data are generated by assuming hormone secretion to occur as a succession of instantaneous release pulses, distributed as Poisson events, separated by quiescent intervals. The pulses are convolved to simulate cumulation of consecutive events and clearance of the hormone. Randomly generated errors, corresponding in magnitude to typical experimental measurement error, are then added to the convolved series. The choice of different values for simulation parameters (e.g., frequency and amplitude of pulses) allows one to emulate some typical physiological patterns of hormone secretion for luteinizing hormone, growth hormone, and thyrotropin or other hormones. Various subsets can be extracted from a simulated time series to study the effect of sampling frequency on the detection of pulses. We show that in sampled series the "observable frequency" of pulses is less than the true nominal frequency. Methods for evaluating pulse-detection algorithms and expressing the results are presented. Simulations of LH secretion were analyzed with the program DETECT. We show that minimizing F+ error rates only might lead to excessively high F- rates. A proper choice of sampling frequency and program probability levels can be made to provide acceptable F+ and F- error rates for various patterns of hormone secretion.