The distribution of fitness effects (DFE) of new mutations is a fundamental parameter in evolutionary genetics. While theoretical models have emphasized the importance of distinct biological factors, such as protein folding, back mutations, species complexity, and mutational robustness at determining the DFE, it remains unclear which of these models can describe the DFE in natural populations. Here, we show that the theoretical models make distinct predictions about how the DFE will differ between species. We further show that humans have a higher proportion of strongly deleterious mutations than Drosophila melanogaster. Comparing four categories of theoretical models, only Fisher's Geometrical Model (FGM) is consistent with our data. FGM assumes that multiple phenotypes are under stabilizing selection, with the number of phenotypes defining a complexity of the organism. It suggests that long-term population size and cost of complexity drive the evolution of the DFE, with many implications for evolutionary and medical genomics.