Molecular biology has shown that most phenotypes are affected by complex, interacting gene networks. Yet, evolutionary biologists and population geneticists often assume very simple genetic architectures where genes act more of less independently of each other. These assumptions are made for conceptual clarity and are partially justified through statistical definitions of gene effects that capture the main effects of alleles as averages over many complex interactions. Still, recent research has revealed many systematic patterns of epistasis and pleiotropy with population genetic effects that should not be ignored. This makes a detailed understanding of genetic architecture important for many questions in evolutionary biology, including the basis of evolvability, and also instrumental for many questions in animal and crop improvement, including our ability to predict selection limits and to avoid unwanted side effects of artificial selection. The philosophy of this project is to develop methods that yield operational measurements of dynamically relevant parameters. A serious problem is the study of genetic architecture has been that many of the parameters in use are not derived from process models, epistatic variance components being the most obvious example. For this reason, it is necessary to combine the statistical models with process models to ensure a general understanding of the meaning of estimators.