Genome-wide association studies (GWASs) have discovered thousands of variants that are associated with human health and disease. Whilst early GWASs have primarily focused on genetically homogeneous populations of European, East Asian and South Asian ancestries, the next-generation genome-wide surveys are starting to pool studies from ethnically diverse populations within a single meta-analysis. However, classical epidemiological strategies for meta-analyses that assume fixed- or random-effects may not be the most suitable approaches to combine GWAS findings as these either confer low statistical power or identify mostly loci where the variants carry homogeneous effect sizes that are present in most of the studies. In a trans-ethnic meta-analysis, it is likely that some genetic loci will exhibit heterogeneous effect sizes across the populations. This may be due to differences in study designs, differences arising from the interactions with other genetic variants, or genuine biological differences attributed to environmental, dietary or lifestyle factors that modulate the influence of the genes. Here we compare different strategies for meta-analyzing GWAS across genetically diverse populations, where we intentionally vary the effect sizes present across the different populations. We subsequently applied the methods that yielded the highest statistical power to a trans-ethnic meta-analysis of seven GWAS in type 2 diabetes, and showed that these methods identified bona fide associations that would otherwise have been missed by the classical strategies.