The durations of caldera-forming eruptions that produce large volumes (> 100 km3) of pyroclastic ejecta are poorly understood due to the absence of direct observations. However, clarifying the timescale of these catastrophic hazardous events is essential for understanding associated eruption dynamics and links with the eruptible portions of the underlying magma system. Case studies addressing the time scale of caldera-forming eruptions are reviewed. Three Quaternary large volume caldera-forming eruption deposits from Yellowstone (US), Taupo (New Zealand) and Kutcharo (Japan) volcanoes are inferred from deposits to have lasted for periods of at least for months to years. However, these estimations are generally based on geological evidence, such as re-worked deposits between eruption units, and not quantitative evidence. Proposed here is a new method of timescale estimation based on paleomagnetic secular variation. A sampling procedure was developed for accurate oriented samples of pyroclastic deposits including volcanic ash. This procedure makes it possible to obtain the mean remanent magnetization of a tephra layer with a 95% confidence limit of about 2°, which is comparable to those of well-determined directions for lava. Based on this procedure, the 7.3 ka Kikai caldera-forming eruption was investigated as a trial. Samples for paleomagnetic measurement were collected from the basal ash-rich part of the lowermost plinian pumice fall (Koya pumice fall) at Satsuma Iwo-jima, Kyushu. The difference of 6.9 degrees in the remanent magnetizations between Koya pumice and reported data of the uppermost co-ignimbrite ash-fall (Kikai–Akahoya Ash) suggests the caldera-forming eruption of a considerable duration (> 50 years) on the basis of average rate of secular variation. Paleomagnetic directions from pyroclastic deposits could be a powerful tool for estimating timescales of large explosive eruptions.