A sire-maternal grandsire threshold model was used for genetic evaluation of stillbirth in US Holsteins. Calving ease and stillbirth records for herds reporting at least 10 dead calves were extracted from the Animal Improvement Programs Laboratory database. About half of the 14 million calving ease records in the database had a known livability score, mostly from herds processed by Dairy Records Management Systems (Raleigh, NC). Calf livability scores of 2 and 3, representing calves born dead and calves that died within 48 h of parturition, respectively, were combined into a single category. The model included effects of herd-year, year-season, parity-sex, sire, birth year group of sire, maternal grandsire (MGS), and birth year group of MGS. Herd-year, sire, and MGS were random effects. Mean predicted transmitting abilities, expressed as the expected percentage of stillbirths, were 7.9 and 8.6 for direct and maternal stillbirths, respectively. Mean reliabilities for both the direct and maternal effects were 45%. Correlations among domestic and Interbull stillbirth solutions on the underlying scale for bulls with at least 90% reliability ranged from 0.63 to 0.90 across countries for direct stillbirths and from 0.69 to 0.96 for maternal stillbirths, indicating that results were generally consistent with those from other countries. There was no evidence of a genetic trend for either trait. More complete recording of stillbirth scores would improve reliabilities and could allow for evaluations of other breeds.