According to currently available seismicity catalogues, seismicity (for example, the number of events with Ms?8) around the turn of the century, from 1897 to 1906, was significantly higher than in recent years. However, the magnitudes of the earthquakes which occurred during this period were determined by Gutenberg, who used the records obtained by the undamped Milne seismograph with the assumption that the effective magnification is 5. Because of saturation of the Milne seismogram for very large events used by Gutenberg for calibration, the gain (=5) used by Gutenberg could have been underestimated, and therefore the magnitude overestimated. Because of the lack of damping, the magnification of this instrument needs to be calibrated carefully. In order to calibrate the instrument response, a Milne seismograph was constructed and has been in operation side by side with damped seismographs at Pasadena. Eleven events have been recorded since February 1977. On the basis of (1) comparison of the amplitudes measured on the Milne seismograms with those of the standard seismograms, (2) numerical experients simulating the response of the Milne seismographs to surface waves, and (3) examination of Gutenberg's original materials used for the calibration, we conclude that the average effective gain is as large as 20 for very large earthquakes, resulting in systematic reduction of the magnitude of up to 0.6. This reduction is large enough to suggest that the turn-of-the-century seismicity peak is of marginal significance.