The results of different algorithms for automatic detection of wave packets in geomagnetic records are compared: the approach based on the Fourier transform in a running time window (F-method), and the approach based on discrete perfect set (DPS) analysis. Using 1-sec data from a mid-latitude INTERMAGNET station we determined with both algorithms such statistical properties of Pc3 geomagnetic pulsations in the frequency range 30-80 mHz as diurnal variations, frequency occurrence, amplitude probability distributions, etc. The diurnal distributions of the occurrence probability determined by both methods have a wide peak around noon hours and a small enhancement during nighttime hours. The advantage of the DPS analysis is the possibility to determine duration of a wave packet. This method shows that average duration of dayside wave packets is statistically longer than that of the nighttime events. Statistical distributions over power/amplitude can be used to infer some conclusions about physical mechanisms of signals. The distribution obtained with the F-method differs for two horizontal components: component distribution is vulnerable to the magnetospheric resonance effect and reveals parabolic dependence in log-log scale, whereas non-resonant component has a power-law distribution. The DPS-method provides almost exponential decay of probability with amplitude for both components. The difference between statistical results of two methods is probably caused by different dependences of detection sensitivity to signal amplitude.