A simple and classical approach to genetic understanding of phenotype variability is to assume that allele influence is additive. However this may be just a mirage of imperfect methodology.
In any case the additive model approach has reached a point when it is quite obvious not enough to explain the genetic background of phenotype heritability.
Gibran Hemani et al., An Evolutionary Perspective on Epistasis and the Missing Heritability. PLoS Genetics 2013. Open access → LINK [doi:10.1371/journal.pgen.1003295]
AbstractThe relative importance between additive and non-additive genetic variance has been widely argued in quantitative genetics. By approaching this question from an evolutionary perspective we show that, while additive variance can be maintained under selection at a low level for some patterns of epistasis, the majority of the genetic variance that will persist is actually non-additive. We propose that one reason that the problem of the “missing heritability” arises is because the additive genetic variation that is estimated to be contributing to the variance of a trait will most likely be an artefact of the non-additive variance that can be maintained over evolutionary time. In addition, it can be shown that even a small reduction in linkage disequilibrium between causal variants and observed SNPs rapidly erodes estimates of epistatic variance, leading to an inflation in the perceived importance of additive effects. We demonstrate that the perception of independent additive effects comprising the majority of the genetic architecture of complex traits is biased upwards and that the search for causal variants in complex traits under selection is potentially underpowered by parameterising for additive effects alone. Given dense SNP panels the detection of causal variants through genome-wide association studies may be improved by searching for epistatic effects explicitly.
It is difficult for me to explain (or even understand well in some aspects) the issues under debate here but some excerpts from the text do ring very true in my mind:
There exists a paradox in evolutionary biology. Despite a near-ubiquitous abundance of genetic variation 
traits under selection often evolve more slowly than expected and,
contrary to expectation, genetic variation is maintained under
selection. This problem is known as ‘stasis’ , , and it is particularly evident in fitness-related traits where the genetic variation tends to be highest  yet there is commonly no observed response to selection at all –.…After hundreds of genome-wide association (GWA) studies 
a picture is emerging where the total genetic variation explained by
variants that have been individually mapped to complex traits is vastly
lower than the amount of genetic variation expected to exist as
estimated from pedigree-based studies, a phenomenon that has come to be
known as the problem of the ‘missing heritability’ .
Again, there are probably numerous contributing factors, and ostensibly
the most parsimonious explanation is that complex traits comprise many
small effects that GWA studies are underpowered to detect , , but whether this is the complete story deserves exploration.…Beyond the realm of complex trait genetics it appears that epistasis
does appear to be common. For example in molecular studies it is routine
to observe ‘phenotypic rescue’ where the phenotypic effect of a gene
knockout can be masked by a second knockout (e.g. ). Another commonly encountered form of epistasis is ‘canalisation’ , where phenotypes exhibit robustness to the knockout of one gene, requiring a second knockout to elicit a response (e.g. ).
At the macroevolutionary scale, epistasis is also of central
importance, for example it has recently been shown that an advantageous
substitution in one species is often found to be deleterious in others,
thus the substition’s effect on fitness is dependent upon the genetic
background in which it is found .…The results suggest that we should expect significant levels of
non-additive variation to be maintained in fitness-related traits.
One of the criticisms seems to suggest that studying homogeneous populations in GWAS is inefficient because linkage disequilibrium (LD) masks the non-additive effect of the alleles, making them appear, erroneously, as additive.
… it was observed that even with modest reductions in LD between causal
variants and observed SNPs all testing strategies tended to decline in
performance rapidly.…The example of the single locus case, overdominance, is central to processes of heterosis and inbreeding depression , , and has been identified in molecular studies also , .
Indeed, heterozygote advantage plays an important role in evolutionary
theory, as it confers segregational load on a population, and this type
of load cannot be purged due to balancing selection, potentially
rendering populations susceptible to accumulating a critical mass of
such polymorphisms .…It is important to note that the processes underlying stasis and missing
heritability are unlikely to be caused by any single factor. For
example, a compelling argument is that though most traits exhibit
genetic variation, selection acts upon multidimensional trait space in
which there is no genetic variation , and this will hold under an additive model of genetic variation.