In the flora of French Guiana we find considerable within-plant variat
ion in leaf form. We observed entire, two-lobed, and three-lobed leave
s within five separate levels (tiers) of the canopy of a single indivi
dual of Pourouma tomentosa subsp. maroniensis. Five branches from each
of the five tiers of the tree were collected around the axis of the t
runk. From these branches five secondary branchlets were selected and
all leaves excised with information recorded as to nodal position, num
ber of leaf nodes, and fertility status of the main branch. This desig
n produced 1015 leaves representing about 20 m(2) of foliar area and a
bout 2.4 kg of blade dry weight. Our objectives were to determine if s
tatistically significant patterns exist for leaf variation and to sugg
est improvements for future, general collections. The four lower tiers
had 62% entire, 10% 2-lobed, and 28% 3-lobed leaves, in contrast to t
he top tier with 38% entire, 11% 2-lobed, and 51% 3-lobed leaves. The
top tier had no fertile branches. In the lower tiers, fertile branches
produced 68% entire leaves whereas nonfertile branches produced only
46% entire leaves. In the top tier, lobed leaves made up 73% of surfac
e area, while in the lower tiers, lobed leaves made up only 48% of tot
al surface area. We selected a random subset of 75 leaves from the 101
5, for morphometric analysis using two-way ANOVA (tier X leaf type). T
he boundaries of leaf images were digitized and rendered into Fourier
coefficients, yielding leaf surface area and two variables that quanti
fy aspects of shape: dissection index and leaf complexity. The Fourier
coefficients were averaged by tier and by leaf type to reconstruct sy
nthetic, average leaf images. Logistic regression was used to predict
the position of leaves on the tree and to provide visualization of the
relationships between leaf position on the tree and leaf morphologica
l variables. Within the tree crown, leaf surface area and leaf specifi
c mass (LSM) increases with height, although leaf shape does not chang
e with height. LSM does not vary with leaf form; and sun leaves are la
rger than shade leaves on this tree. We conducted computer sampling ex
periments based on exact randomization to simulate the process of obta
ining all leaf shapes present in an individual tree when making field
collections of varying numbers of duplicates. This also points out the
importance of noting the presence of within-tree variation in leaf fo
rm on herbarium labels. Failure to recognize led variation can lead to
incorrect delimitation of species as well as cause overestimates of t
he number of species in diversity studies.