To obtain quantitative information about population dynamics from counts of animals, the per capita detectabilities of each species must remain constant over the course of monitoring. We characterized lizard detection constancy for four species over 17 yr from a single site in northern Guam, a relatively benign situation because detection was relatively easy and we were able to hold constant the site, habitat type, species, season, and sampling method. We monitored two species of diurnal terrestrial skinks (Carlia ailanpalai [Curious Skink], Emoia caeruleocauda [Pacific Bluetailed Skink]) using glueboards placed on the ground in the shade for 3 h on rainless mornings, yielding 10,286 skink captures. We additionally monitored two species of nocturnal arboreal geckos (Hemidactylus frenatus [Common House Gecko]; Lepidodactylus lugubris [Mourning Gecko]) on the basis of 15,212 sightings. We compared these count samples to a series of complete censuses we conducted from four or more total removal plots (everything removed to mineral soil) totaling 400 m² (about 1% of study site) in each of the years 1995, 1999, and 2012, providing time-stamped quantification of detectability for each species. Unfortunately, the actual population trajectories taken by the four species were masked by unexplained variation in detectability. This observation of debilitating latent variability in lizard detectability under nearly ideal conditions undercuts our trust in population estimation techniques that fail to quantify venue-specific detectability, rely on pooled detection probability estimates, or assume that modulation in predefined environmental covariates suffices for estimating detectability.