Abstract Introduction A recent study found a significant increase of ABCA7 loss-of-function variants in Alzheimer’s disease (AD) cases compared to controls. Some variants were located on noncoding regions, but it was demonstrated that they affect splicing. Here, we try to replicate the association between AD risk and ABCA7 loss-of-function variants at both the single-variant and gene level in a large and well-characterized European American dataset. Methods We genotyped the GWAS common variant and four rare variants previously reported for ABCA7 in 3476 European–Americans. Results We were not able to replicate the association at the single-variant level, likely due to a lower effect size on the European American population which led to limited statistical power. However, we did replicate the association at the gene level; we found a significant enrichment of ABCA7 loss-of-function variants in AD cases compared to controls (P = 0.0388; odds ratio =1.54). We also confirmed that the association of the loss-of-function variants is independent of the previously reported genome-wide association study signal. Conclusions Although the effect size for the association of ABCA7 loss-of-function variants with AD risk is lower in our study (odds ratio = 1.54) compared to the original report (odds ratio = 2.2), the replication of the findings of the original report provides a stronger foundation for future functional applications. The data indicate that different independent signals that modify risk for complex traits may exist on the same locus. Additionally, our results suggest that replication of rare-variant studies should be performed at the gene level rather than focusing on a single variant.