Genome-wide association (GWA) studies have been remarkably successful in identifying genetic variants associated with common, complex diseases. In the four and a half years since the advent of this technology, over 450 publications have documented over 500 associations with a significance less than 5 x 10-8 in over 130 diseases and traits. In cancer alone, over 70 well-replicated associations with at least 18 distinct forms of neoplasia have been reported. Many of these variants lie in genes previously unsuspected of being related to the associated conditions, and nearly half are in regions containing no known genes at all. Many variants are associated with seemingly unrelated diseases (Table). These findings are shedding new light on the pathophysiology of complex diseases, as well as identifying promising targets for drug development or variants related to drug selection or dosing. Markers assayed on GWA platforms are likely not causal variants, but instead are only in linkage disequilbrium (inherited together) with them. Identifying an association is thus only a first step, albeit an important one, in determining the change(s) in genomic structure and function underlying a given disease. The many hundreds of thousands of variants assayed in each study are known present a substantial possiblity of spurious, false-positive findings. Replication of associations in different study groups and even in different populations is thus essential in isolating GWA signals worthy of in-depth investigation through techniques such as fine mapping, sequencing, and functional studies. Examination of large numbers of cases with a variety of environmental exposures is necessary for identifying possible gene-environment interactions. Such studies are facilitated by wide availability of GWA data in well-characterized, population-based cohorts with extensive exposure information. The challenge of relating millions of genetic variants to potentially hundreds or thousands of traits and exposures, though computationally and inferentially daunting, is essential to understanding the joint role of genes and environment in complex diseases such as cancer. Once a functional variant is identified, its value in risk assessment, disease detection, and treatment selection must be assessed in translational studies for it to be useful clinically. Identifying available therapies, or developing novel treatments to modify the effects of a causal variant, typically takes even greater effort. Surprisingly, most GWA-defined variants confer only small increased risks of disease (less than 30% increase) and taken together generally explain only a small proportion of the familial clustering of a given trait, leaving a large fraction unexplained. Promising approaches for finding this “missing heritability” include targeted or whole-genome sequencing to identify rarer variants (present in less than 5% of the population) that may be in linkage disequilibrium with the typically more common variants assayed on GWA platforms. Identification of even a few dozen low-frequency (0.5%-5%) variants with moderate effects (two- to three-fold odds ratios of disease), would suffice to account for the unexplained heritability in type 2 diabetes, for example. Other promising methods include exploring GWA data for structural variation such as duplications or deletions of germline DNA, and for evidence of interaction among genes. Expanding GWA sample sizes through meta-analyses and consortia, especially in persons of non-European ancestry, also has enormous value in finding variants of small effect and breaking up genomic regions of linkage disequilibrium to better focus the search for causal variation. The value of existing and future GWA studies for identifying disease-related variants can best be enhanced by ensuring wide availability of high-quality data to qualified researchers with appropriate protections for consent and privacy; increasing sample sizes and diversity and ensuring meticulous meta-analyses, particularly for conditions with relatively small sample sizes studied to date; and enhancing phenotyping to include subtler or more quantitative or precise phenotypes. Improvement of GWA platforms should permit capture of larger proportions of variation in implicated genes and expand investigation of the X chromosome, particularly as imputation of X and Y markers improves. Rigorous measurement of environmental exposures, including rare exposures in common diseases, and epigenetic variants in appropriate tissues when technically feasible, will facilitate identification of gene-environment interactions. And although structural variation is at present difficult to assay, linkage disequilibrium patterns in GWA data and improved maps can be used to identify common structural variants reliably. High-throughput sequencing techniques and programs such as the 1,000 Genomes Project will expand identification and tagging of structural variants even further. All of these steps will lead GWA studies into critical new directions of defining the genomic structural and functional changes underlying complex diseases and approaches for modifying their effects. Citation Format: Teri Manolio. Genetic architecture of cancer and other complex diseases: Lessons learned and future directions [abstract]. In: Proceedings of the AACR 101st Annual Meeting 2010; 2010 Apr 17-21; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2010;70(8 Suppl):Abstract nr SY32-02