We devised a new method to estimate the density of primate groups in habitats that preclude the use of a line-transect census because the ground is too steep. We combined point census and group follows. From the number of groups counted at a fixed point for an hour, n, group density D was calculated: D = lambda n / pi. Lambda, the detectability constant, was a constant when distance-dependent detectability g(y) was regressed on a half-normal model: g(y) = e (-lambda y(2)) and can be estimated by combining the information of group follow and point census. Using this method, we estimated the group density of Japanese macaques in Yakushima. A census area of 7 km(2) was divided into 28 grid squares (500 m x 500 m). One observer was positioned at a point in each grid square, and those points were censused simultaneously for 4-6 days from 0600-0700 to 1500-1600 hr. Four troops were followed for 144 hr during the point census. Distance-dependent detectability closely correlated with the half-normal model. The detectability constant varied with the time of day, but it was not influenced by troop identity or topography. Group density was calculated to be 1.48 +/- 0.61 and 0.701 +/- 0.432 groups/km(2) in the disturbed and undisturbed areas, respectively (95% confidence limit). "True" group density estimated by home range data was within the confidence limit calculated by a point census in the home range of the troops for two troops, suggesting that this method was valid. This method is applicable to other species as long as at least one group can be followed, because it satisfies the fundamental assumptions of point census, and the detectability does not seem to be biased by troop or topography.