Assessment of forest health on a national scale is a difficult task. One pa
rameter used to rapidly classify tree health is crown transparency, or the
amount of sky seen through the crown. Previous investigations of sources of
error in estimating crown transparency concentrated on observer experience
and perception, and on lighting conditions, showing that the error can be
as much as +/- 15%. Here we tested the hypothesis that crown dimensions, wh
ich determine the path-length of the line of sight of the observer through
crowns, introduce a large bias in estimates of crown transparency. Both the
oretical and empirical results show that crown transparency is highly sensi
tive to crown dimensions. Crowns of trees with path-lengths > 10 m are alwa
ys likely to be rated < 30% transparent (i.e., considered healthy), althoug
h their crowns may be as unhealthy as those of trees with path-lengths <4 m
, and rated > 80% transparent. The misclassification of tree health is furt
her exacerbated by the reduction in the rate at which transparency decrease
s per unit of path-length as the path-length increases, reflecting a lower
average leaf density with increasing crown size. Thus we propose that avail
able crown transparency data may be used to rank relative tree health withi
n narrow intervals of path-length, thereby incorporating the effects of cro
wn dimensions on transparency.