Abstract. There are many sources of variability in gene-phenotype associations. During the measurement of genotype and phenotype and during selection, researchers must deal with experimental error in trials; gene-gene interaction (epistasis) for sub-traits and observed traits; trait-trait interaction (pleiotropy) and gene-or genotype-by-environment interaction. These effects can be structured in a framework that allows simulation of the entire gene-environment 'landscape'. Studies of these landscapes have been published by others. Here we aim to explain with simple examples some of the types of insights that can be made. A current challenge for breeders working with simple marker-phenotype associations is to design selection strategies that can rapidly create new combinations of multiple marker-based traits. For a real-world example in wheat, we have used simulation to show how gene enrichment during early generations (selection of homozygotes and heterozygotes with desirable alleles) can greatly reduce resource requirements when combining 9 genes into one genotype through marker-assisted selection. Another wheat example compares phenotypic and QTL-based selection for coleoptile length where the QTL also had a pleiotropic association with plant height. These simulations show the relative negative effects of either low heritability, or less than complete detection of QTL associated with traits. Finally, we revisit a marker-assisted selection (MAS) example whereby a QTL study is undertaken on a population for a complex trait, and then those QTL are used in selection. This process is subject to all sources of error described above. If the trait is complex, then interactions among sub-traits; between sub-traits and the environment; or between the chromosomal locations of controlling genes, create an extremely 'rugged' selection landscape that slows breeding progress. In this situation, a detailed understanding of some of these interactions is required if MAS is to be able to exceed the progress of conventional breeding.