Background Gene-panel testing for breast cancer susceptibility is widely used, but for many genes, the evidence for association with breast cancer is weak, underlying risk estimates are imprecise, and reliable subtype-specific risks are lacking. Methods We performed gene panel sequencing for 34 putative susceptibility genes on samples from 60,466 female breast cancer cases and 53,461 controls. We estimated odds ratios (ORs) for overall breast cancer and tumor subtypes, separately for protein truncating and rare missense variants, in aggregate. We evaluated missense-variant associations by domain and according to classification of pathogenicity. Results Protein-truncating variants were associated with overall breast-cancer risk at P<0.0001 for five genes: ATM, BRCA1, BRCA2, CHEK2 and PALB2. Associations between risk and truncating variants of four other genes (BARD1, RAD51C, RAD51D and TP53) were significant (P<0.05) and had Bayesian false discovery probabilities <5%. The upper 95% confidence limit excluded a twofold risk for overall breast cancer for truncating variants of 19 of the remaining 25 genes. ORs were higher for estrogen receptor (ER)-positive disease for CHEK2 and ATM, and higher for ER-negative disease for BARD1, BRCA1, BRCA2, PALB2, RAD51C, RAD51D and TP53. Rare missense variants in CHEK2, ATM and TP53, in aggregate, were associated with risk (P<0.001). For BRCA1, BRCA2 and TP53, missense variants classified as pathogenic by standard criteria were associated in aggregate with risk, the risk being similar to that of truncating variants. Conclusions These results define the genes most clinically useful for inclusion on panels for breast cancer risk prediction, and estimates of the risks associated with truncating variants, to guide genetic counselling.