Modular growth generates at least three kinds of hierarchies: morpholo
gical, functional, and demographic. The morphological hierarchy corres
ponds to the phenotypic characters of both the units (modules) that ar
e repeated by developmental processes, and the units (organisms, colon
ies, and clones) that develop by iteration, specialization and integra
tion of modules. The functional hierarchy concerns the levels of inter
action or, in evolutionary terms, the functional relationships between
fitness and the phenotypic characters at different levels of modular
organization. Finally, clonal growth and reproduction results in a nes
ted hierarchy of demographic units that are replicated by asexual prop
agation. Each level of the demographic hierarchy that is characterized
by specific birth and death rate, is a potential candidate for evalua
ting fitness. We propose a formal approach in order to analyze the hie
rarchical structure of phenotypic selection in modular organisms, and
to evaluate the selective importance of various levels of modular orga
nization. We derive a measure of selective importance from the sensiti
vity of fitness to a unit change in the characters of a given level: t
he sum of squared sensitivities associated with that level. We propose
that clonal-level characters of disintegrated clones will make small
contributions to the variation in fitness, while such characters will
be more important if the clone is physically and physiologically integ
rated. Moreover, we present a decomposition of fitness variation in re
lation to the levels of trait variation. This decomposition demonstrat
es that the levels where variation in fitness is observed do not alway
s correspond to the interaction-levels at which the causal agents of s
election are acting on particular traits. Following the logic of pheno
typic optimization models, we consider three examples of selection in
order to examine whether the different demographic levels are equally
suitable for evaluating fitness. In two examples of density-independen
t selection we show that the Malthusian parameter is identical at all
levels in the hierarchy. However, the third example shows that this re
sult is not valid in density-dependent selection models. The way densi
ty-dependent regulation is supposed to operate in the model system det
ermines which of the demographic levels should be used to evaluate fit
ness. Consequently, there is no fundamental demographic level that sho
uld a priori be chosen when measuring fitness.