Recently, the bunching technique has been widely utilized in the Feynman-alpha experiment using a multi-channel scaler (MCS) to measure the prompt neutron decay constant alpha(p). Although the bunching technique enables us to perform efficient experiments, it was pointed out that an inherent count-loss process arises due to the channel advance time Delta between adjacent MCS channels. Through derivation of a Feynman-alpha variance-to-mean formula containing Delta by means of the multi-gate PBI-Bell equation, Yamane and Hayashi ([Yamane, Y. & Hayashi, Y. 1995]. Annals of Nuclear Energy, 22(8), 533) indicated that this count-loss process does not play any important roles when the channel advance time is much smaller than the dwell time T. However, the Delta/T ratio often becomes large in thermal systems at deep-subcritical states or fast ones, because the dwell time should be chosen to be much smaller than reciprocals of alpha(p) Values for such systems. On the other hand, since the ratio of the dead time d of neutron detectors to the dwell time becomes also large when the Delta\T ratio is not small, the count-loss process due to the dead time cannot be neglected. Therefore, Feynman-alpha variance-to-mean and covariance-to-mean formulae containing both Delta and d were derived by means of the compound detection probabilities. Based on the covariance-to-mean formula, a new experimental technique was developed and examined at the Kyoto University Critical Assembly. The result of the examination indicated that one can measure exact alpha(p) values when Delta/T ratios are known, even though Delta/T and d/T ratios are not small. (C) 2000 Elsevier Science Ltd. All rights reserved.