Notwithstanding this defect, personal impressions of an earthquake-shock give a fairly accurate, if incomplete, idea of its nature. Nearly all observers placed under favourable conditions agree that an earthquake begins with a deep rumbling sound, accompanied, after the first second or two, by a faint tremor which gradually, and sometimes rapidly, increases in strength until it merges into the shock proper, which consists of several or many vibrations of larger amplitude and longer period, and during which the attendant sound is generally at its loudest; the earthquake dying away, as it began, with tremors and a low rumbling sound. [Illustration: FIG. 79.--Seismographic Record of Tokio Earthquake of 1894. (_Omori._)] The vibrations that produce the sensible shock are by no means all that are present during an earthquake. The Indian earthquake, for instance, seemed to last about three or four minutes at Midnapur; but the movements of the bubble of a level showed that the ground continued to oscillate for at least five minutes longer (p. 280). Many of these unfelt waves are rendered manifest by seismographs, although there are still others that elude registration either from the extreme shortness or the great length of their periods. In Fig. 79 is shown the principal part of a diagram obtained at Tokio during the Japanese earthquake of June 20th, 1894 (p. 18), the curve representing the N.E.-S.W. component of the horizontal motion during the first 25 seconds of the record. The instrument employed is one specially designed for registering strong earthquakes, and is unaffected by very minute tremors. Those which formed the commencement of this earthquake lasted for about 10 seconds, as shown by ordinary seismographs, and the vibrations had attained a range of a few millimetres before they affected the instrument in question. For the first 2-1/2 seconds, they occurred at the rate of four or five a second. The motion then suddenly became violent, and the ground was displaced 37 mm. in one direction, followed by a return movement of 73 mm., and this again by one of 42 mm., the complete period of the oscillation being 1.8 seconds. The succeeding vibrations were of smaller amplitude and generally of shorter period for a minute and a half, then dying out during the last three minutes as almost imperceptible waves with a period of two or more seconds.[81]