The term seismograph is today reserved for instruments recording the waveform of the ground motion versus time. In that sense, the first seismograph was built in Italy by Cecchi in 1875; it was however so unsensitive that its first known seismogram dates from 1887. In meantime, Ewing and colleagues in Japan had built several seismographs and recorded the first earthquake in 1880 [Ewing 1884]. One of the instruments was in the same year tested on a shake table! Von Rebeur-Paschwitz [Von Rebeur-Paschwitz 1889] recognized seismic waves from an earthquake in Japan in the records of his tiltmeters at Potsdam and Wilhelmshaven, giving seismology a global dimension. The early history of seismometry to 1900 is described by Dewey and Byerly [Dewey & Byerly 1969] in an excellent article with many figures and references.
Going back in time, the "electromagnetic seismograph" built by Palmieri in 1856 was little short of being the first seismograph in a modern sense. It had motion-sensitive electric contacts whose closures were recorded on a strip of paper like Morse code. Earlier constructions, which were only designed to indicate the occurrence and direction of a seismic shock, would today be termed seismoscopes. The Chinese Chang Heng is reported to have built one in 132; models of his jar-shaped instrument are exhibited in many seismological institutes but its inner mechanism is unknown.
In the beginning of the 20th century, the technical development concentrated on mechanical seismographs with smoke-paper recording. Viscous damping was introduced by Wiechert around 1900. To overcome the remaining solid friction, the mass had to be increased with the square of the magnification. The largest seismographs had masses from 10 to 20 tons, magnifications around 1000, and stable free periods up to 12 s. Mainka and Wiechert seismographs served in many obervatories until after the second world war, and a few of them are still (or again) operational. De Quervain and Piccard in Zürich [de Quervain & Piccard 1924,de Quervain & Piccard 1927] built a mechanical three-component seismograph with a single mass of 21 tons whose position was stabilized with a water ballast - probably the first feedback-stabilized seismograph. Many of the seismographs of the early 20th century are described in Galitzin's lectures on seismometry [Galitzin 1914] and in a comprehensive handbook article by Berlage [Berlage Jr. 1932].
Photographic recording was occasionally used from the beginning but the higher cost and lower quality of the record put the method at a disadvantage, at least until electric light was available. Later it became a practical alternative, for example with the Wood-Anderson torsion seismograph on which the Richter magnitude scale is based [Anderson & Wood 1925]. The electromagnetic seismograph with galvanometer-photopaper recording, invented by Galitzin already in 1904, remained for more than half a century the most sensitive long-period seismograph but had to wait for gradual improvements by LaCoste, Benioff, Press, Ewing, and Lehner [LaCoste 1934,Benioff & Press 1958,Press et al. 1958,Lehner 1959] before it was stable enough for wide deployment in the WWSSN (Worldwide Standardized Seismograph Network [Oliver & Murphy 1971].
The next generation of electromagnetic seismographs in the HGLP (High-Gain Long-Period) project [Savino et al. 1972] was partially electronic, using galvanometer-phototube amplifiers. The Seismic Research Observatory (SRO) had a fully electronic, broadband, force-balance sensor but did not record the broadband signal [Peterson et al. 1976]). The sensor of the original IDA network (International Deployment of Accelerometers) was a LaCoste-Romberg gravimeter with a slow electrostatic force-balance feedback [Agnew et al. 1976,Agnew et al. 1986]; although this instrument was not useful as a general-purpose seismometer, its sensitivity in the free-mode band is unsurpassed. An eyewitness account of the emerging electronic era of seismometry from 1947 on was given by Melton [Melton 1981a,Melton 1981b].
In the time of transition from electromagnetic to electronic seismographs between 1960 and 1975, two opposite trends can be observed. As long as visible recording was the standard and magnetic tape recording was not much better, the gain could only be increased when the marine microseisms, at periods around 6 s, were suppressed. This resulted in the development of high-gain, narrow-band seismographs which were excellent for studying ground noise and monitoring nuclear explosions but were easily saturated by earthquakes. On the other hand, several broadband seismographs with analog or digital magnetic tape recording were developed. They remained experimental because continuous broadband recording and digital or analog post-processing were too inconvenient for routine work. The first digital broad-band seismograph was operated at CALTECH as early as 1962 [Miller 1963] with the intention ``to preserve the greatest spectrum, dynamic range, and sensitivity''. The installation was discontinued because the digital technology was too inefficient at the time. Block and Moore [Block & Moore 1970] built a small broadband quartz accelerometer which was the most sensitive broadband sensor of its time but not a very practical instrument; it required vacuum and a thermostat. An analog very-broad-band seismograph was operated in Czechoslovakia from 1972 on [Plesinger & Horalek 1976]; its data archive was later converted to a digital standard format. The first practically successful digital broadband installation is the German GRF array [Harjes & Seidl 1978,Buttkus 1986] which has been operational since 1976. The present generation of digital very-broad-band seismographs covering the full teleseismic bandwidth including the free-mode band was developed from 1984 on [Wielandt & Steim 1986].