Genome‐wide association studies (GWAS) that allow for allelic heterogeneity may facilitate the discovery of novel genes not detectable by models that require replication of a single variant site. One strategy to accomplish this is to focus on genes rather than markers as units of association, and so potentially capture a spectrum of causal alleles that differ across populations. Here, we conducted a GWAS of Alzheimer disease (AD) in 2,586 Swedes and performed gene‐based meta‐analysis with three additional studies from France, Canada, and the United States, in total encompassing 4,259 cases and 8,284 controls. Implementing a newly designed gene‐based algorithm, we identified two loci apart from the region around APOE that achieved study‐wide significance in combined samples, the strongest finding being for FRMD6 on chromosome 14q (P = 2.6 × 10−14) and a weaker signal for NARS2 that is immediately adjacent to GAB2 on chromosome 11q (P = 7.8 × 10−9). Ontology‐based pathway analyses revealed significant enrichment of genes involved in glycosylation. Results suggest that gene‐based approaches that accommodate allelic heterogeneity in GWAS can provide a complementary avenue for gene discovery and may help to explain a portion of the missing heritability not detectable with single nucleotide polymorphisms (SNPs) derived from marker‐specific meta‐analysis. Hum Mutat 33:521–529, 2012. © 2011 Wiley Periodicals, Inc.