Partitioning genetic variance by sets of randomly sampled genes for complex traits in D. melanogaster and B. taurus, has revealed that population structure can affect variance decomposition. In fruit flies, we found that a high likelihood ratio is correlated with a high proportion of explained genetic variance. However, in Holstein cattle, a group of genes that explained close to none of the genetic variance could also have a high likelihood ratio. This is still a good separation of signal and noise, but instead of capturing the genetic signal in the marker set being tested, we are instead capturing pure noise. Therefore it is necessary to use both criteria, high likelihood ratio in favor of a more complex genetic model and proportion of genetic variance explained, to identify biologically important gene groups