A previous genome-wide association study (GWAS) identified common variation at the BARD1 locus as being highly associated with susceptibility to high-risk neuroblastoma, but the mechanisms underlying this association have been not extensively investigated. Here, we performed a fine mapping analysis of BARD1 locus (2q35) using GWAS data from 556 high-risk neuroblastoma patients and 2,575 controls of European-American ancestry, and identified two independent genome-wide neuroblastoma-associated loci. Functional single-nucleotide polymorphism (SNP) prioritization identified two causative variants that independently contributed to neuroblastoma risk, and each replicated robustly in multiple independent cohorts comprising 445 high-risk cases and 3,170 controls (rs17489363: combined p = 1.07 × 10−31, OR:1.79, 95% CI:1.62–1.98 and rs1048108: combined p = 7.27 × 10−14, OR:0.65, 95% CI:0.58–0.73). Particularly, the T risk allele of rs17489363 in the canonical promoter region of full-length BARD1 altered binding site of the transcription factor HSF1 and correlated with low expression of full-length BARD1 mRNA and protein. Low-level expression of full-length BARD1 associated with advanced neuroblastoma. In human neuroblastoma cells, attenuating full-length BARD1 increased proliferation and invasion capacity. In conclusion, we have identified two potentially causative SNPs at the BARD1 locus associated with predisposition to high-risk neuroblastoma, and have shown that full-length BARD1 may act as tumor suppressor.