Fragile X syndrome is rare but a prominent cause of intellectual disability. It is usually caused by a de novo mutation that occurs on multiple haplotypes and thus would not be expected to be detectible using genome-wide association (GWA). We conducted GWA in 89 male FXS cases and 266 male controls, and detected multiple genome-wide significant signals near FMR1 (odds ratio = 8.10, P = 2.5 × 10-10). These findings withstood robust attempts at falsification. Fine-mapping yielded a minimum P = 1.13 × 10-14, but did not narrow the interval. Comprehensive functional genomic integration did not provide a mechanistic hypothesis. Controls carrying a risk haplotype had significantly longer FMR1 CGG repeats than controls with the protective haplotype (P = 4.75 × 10-5), which may predispose toward increases in CGG number to the premutation range over many generations. This is a salutary reminder of the complexity of even "simple" monogenetic disorders.