For a study of long-distance migrants in sub-Saharan Africa, a census method was developed that combined precision and accuracy regarding bird numbers and tree choice. The number of birds present in trees and shrubs can be counted accurately, although it is time-consuming. We describe how much time is needed to detect all birds present in trees, using data collected in over 2000 plots across West Africa during the dry season (October–March in 2007–2015). The observation time per tree depended on tree size, number of birds present and the opacity of the crown. The giving-up time of the observers increased with canopy volume, but was independent of the number of birds in a tree. Detection probabilities of bird species differed relative to microhabitat choice and feeding techniques. Species-specific detectabilities hardly varied during the day or the season. All foraging birds and immobile birds (save a few percent in dense canopies) were detected using the individual-tree-approach. Bird density is expressed as number per canopy volume, but little information is lost when density is given as number per canopy surface. The variation in bird density was large and differed per tree species. Within tree species, bird density was related to the opacity of the crown, the abundance of insects and whether there were berries or flowers. These findings suggest that, to collect biologically relevant information, the density of tree-dwelling birds should be measured at the level of the individual tree, and not per surface area, habitat type or tree species (as is typical in published studies).