The purpose of this study was to investigate frass production of the beech caterpillar, Quadricalcarifera punctatella (Motschulsky) (Lep., Notodontidae) and to develop a method for estimating the field density of its last instar larvae from fallen pellets. About 90% of larval frass by weight was produced during the last instar period. The number of pellets falling to the forest floor was a better parameter for density estimation than weight. Pellets with a width of more than 1.8 mm were considered to have been produced during the last instar. The number of pellets produced per day or per hour in the last instar was influenced by many factors, especially by temperature; however, the number of pellets produced during the last instar was relatively constant (ca. 630) regardless of temperature and photoperiod. The number of last instar pellets dropping per m² during the season was measured in traps and divided by 630 to estimate the density of the last instars. This method, based on the Southwood-Jepson graphical method, is very simple but reliable because the estimates are not influenced by ambient temperature, and because the total number of falling pellets can be estimated correctly by setting traps throughout the entire season. Although some frass was destroyed and washed away from the trap, we concluded that no compensation was necessary for field application since the decrease was minimal. The density was underestimated when severe defoliation occurred. In such a situation, the density (Y) should be estimated from the following equation: Y = 0.9881log10 X + 0.898, where X is the maxium value of the frass drop per day (g･ dw/m²/day). Only five traps (1 m² each) were required with a 0.2 ratio of tolerance limit at the 70% confidence level. Such a small number of samples were necessary because the method used counts of frass pellets rather than individuals; larval mobility and wind homogenized the spatial distribution of falling frass.