Hollows in standing trees are an important ecological resource for many Australian vertebrates, including a range of threatened mammals, reptiles and birds. However, the ecology of tree hollows, and the extent to which they support hollow-dependent fauna, has been severely under-studied in northern Australia. This study evaluated the reliability of ground-based surveys of tree hollows in a northern Australian eucalypt savanna. We counted hollows in trees using two approaches: (1) ground-based surveys and (2) climbing surveys, i.e. double sampling. Assuming that climbing surveys are accurate, ground-based surveys tended to overestimate hollow density (per tree) by around 15%. However, of the hollows counted in the climbing surveys (n = 202), 44.5% were blocked by hard termitaria. Once unavailable (blocked) hollows were accounted for, ground-based surveys tended to overestimate available hollow density (per tree) by 59%. According to generalised linear mixed models (GLMMs), a hollow was more likely to be blocked if it had a smaller entrance diameter, occurred on a tree with a termite mound at its base and smaller DBH. We calibrated site-level ground-based counts using GLMMs of climbing counts and available climbing counts as a function of ground-based counts for double-sampled trees. Overall, our uncalibrated counts resulted in over-reporting of mean site-based density of available hollows by 38%. The discrepancy between densities of total hollows and available hollows has important implications for the conservation of many threatened and declining species known to rely on tree hollows in northern Australia. We suggest that when conducting ground-based hollow surveys to assess tree hollow availability in northern Australia, a subsample of trees should be double-sampled, i.e. both ground-based and climbing counts, to account for survey error due to the blocking of hollows by termitaria.