From the DME, note the distance travelled in 36 seconds. Since 36 seconds = 1% of one hour, multiply this figure by 100 to get the groundspeed.
I seem to remember that this is particularly well-suited to fairly fast a/c (i.e. > 250 kts), presumably because the absolute error in speed is less significant, as is the absolute error caused by the difference in slant range -vs- track miles.