The mb : h4, criterion is now generally accepted as the best method of distinguishing on seismic evidence between earthquakes and underground explosions (see for example SIPRI, 1968; Marshall & Basham 1972): for explosions the body wavemagnitude (&) is usually 1-2 magnitude units greater than the surface wave magnitude (Ad,); for earthquakes mb and M , are more nearly equal. Some earthquakes (known to be earthquakes because of their focal depth: see below) however fail on the mb: M , criterion (Landers 1972; CCD 1972) and in this paper we suggest reasons why these earthquakes fail and also why paradoxically these earthquakes may turn out to be the seismic events that are easiest to identify. Fig. 1 (a) shows the P signals from such an earthquake with epicentre near Alma Ata, Kazakhstan, as recorded at Yellowknife, Canada (YKA). The United States Coast and Geodetic Survey (USCGS) give an average mb for this earthquake of 4.9 (mb measured at YKA is 5*0), M, measured, using the methods of Marshall & Basham (1972), from the World Wide Standard Station Network (WWSSN) is 3.7. On the nib : M, criterion using the results of Marshall & Basham (1972) this earthquake lies close to the explosion population. The signal consists of two clear arrivals (which is typical of this type of anomalous earthquake) and at first sight might appear to indicate two explosions fired six seconds apart. Closer examination of the signal however shows the second arrival is of opposite polarity to the first and is thus probably the surface reflection p P . Confirmation that the second pulse is of reversed polarity is given in Fig. 1 (b) where the effects of anelastic attenuation and recording instrument have been removed from the recorded signal using the spike filtering technique described by Douglas et al. (1972b). The record now becomes simply a positive impulse (P) followed by a negative impulse (pP) . The P-pP time appears to be about 6 s indicating a depth of focus of about 20km. The event is thus conclusively identified as an earthquake. To construct a spike filter some value has to be assumed for t* where f * = T J Q A v ; T is the total travel time and QAV is the average quality factor Q, along the path Alma Ata to YKA. T is well known from travel-time studies so that in assuming a vdue oft* we have effectively to assume a value for Q A p If the correct value of QAV is used the spiked seisniogram will show the shape of the P and p P pulses (and any other prominent pulse) at source. If now we assume these pulses at source are either wholly positive or wholly negative (i.e. they do not oscillate about the zero baseline) we have a way of estimating QAv: we simply spike the record with a range of values of t* and select the value that gives a spiked seismogram with pulses that do not oscillate about the baseline. (To obtain pulses that oscillate requires that the earthquake source oscillates and the theoretical studies of Burridge (1968) using an earthquake model of slip on a plane shows that any oscillations are likely to be negligible where friction resists slipping.) From Fig. 1 for example we see that using a value of t* = 0.4 gives P and p P pulses that swing strongly positive and negative whereas for I* = 0.2, P and p P are almost simple impulses. Even with f * = 0.2 however the P pulse seems to have a small overshoot. Using smaller values of f * reduces the