Genetic-by-environmental interactions were estimated from national data by separately adding random regressions for heat stress (HS) and herd yield level (HL) to the US all-breed animal model to improve predictions of future records and genetic rankings in other climates and production situations. Yield data included 79 million lactation records of 40 million cows; somatic cell score, productive life, and daughter pregnancy rate were also tested but had fewer records. Coefficients for HS were the state’s July average temperature-humidity index; coefficients for HL were management-level weighted means for energy-corrected milk (ECM) divided by breed-year mean ECM. Coefficients were standardized to a mean of 0 and variance of 1. Predictions of current (August 2014) from historical (August 2011) records were tested with a model that included herd management group (absorbed), sire estimated breeding value (EBV), dam EBV, and an interaction term (HS or HL) from the truncated data; records were weighted by lactation length for records in progress and by herd heritability using the same weights as in national evaluations. Estimated regression coefficients for sire EBV and dam EBV were always near their expected values of 0.5 and did not change when HS or HL interactions were added to the model. Estimated regressions for interaction terms, expected to be near 1, were 0.80 to 0.93 for HS and 0.61 to 0.72 for HL in yield traits. Squared correlations increased by < 0.0003 for both HS and HL; increases for nonyield traits were even smaller. An additional test used multitrait across-country EBV to predict rankings of the same bulls in the United States and 14 other countries with somewhat different environments. The HS coefficient was significant (P < 0.05) in 9 of 14 countries for milk and protein and in 10 for fat; the HL coefficient was significant in 8 countries for milk, 5 for protein, and 1 for fat. Squared correlations after adding an interaction term increased by < 0.004 for HL and < 0.01 for HS. The small changes in rank and correlation gains when HS and HL interactions were included in national evaluations indicate that current genetic predictions perform very well in a variety of environments.