Several studies have investigated the role of the neuronal sortilin-related receptor (SORL1) gene in Alzheimer's disease (AD), but findings have been inconsistent. We conducted a study of 7 single nucleotide polymorphisms (SNPs), rs668387, rs689021, rs641120, rs1699102, rs3824968, rs2282649, and rs1010159, in the SORL1 gene that were associated to AD in previous studies. We tested for association with AD and cognitive function in 6741 participants of the Rotterdam Study and in 2883 individuals from the Erasmus Rucphen Family study. We performed meta-analyses on AD using our data together with those of previous studies published prior to September 2008 in Caucasians. Further, we studied up to 76 SNPs in a 400 kb region within and flanking the gene to evaluate the evidence that other genetic variants are associated with AD or cognitive function. There was no significant evidence for association between SORL1 SNPs and incident AD patients in the Rotterdam Study. In a meta-analysis of our data with those of others, six out of seven SNPs attained borderline significance. However, removal of the first study reporting association from the meta-analysis resulted in non-significant odds ratios for all SNPs. SNPs rs668387, rs689021, and rs641120 were associated with cognitive function in non-demented individuals at borderline statistical significance in two independent Dutch cohorts, but in the opposite direction. Testing for association using dense SNPs in the SORL1 gene did not reveal significant association with AD, or with cognitive function when adjusting for multiple testing. In conclusion, our data do not support the hypothesis that genetic variants in SORL1 are related to the risk of AD.