Long-term response of genomic selection can be improved by considering allele frequencies of selected markers or quantitative trait loci (QTLs). A previous formula to weight allele frequency of favorable minor alleles was tested, and 2 new formulas were developed. The previous formula used nonlinear weights based on square root of frequency of the favorable allele. The new formulas included a parameter δ to balance long- and short-term progress; one used square root and the other used simple linear weights. The formulas were tested by simulation of 20 generations (population size of 3,000 for each generation) with direct selection on 3,000 QTLs (100 per chromosome). A QTL distribution with normally distributed allele effects and a heavy-tailed distribution were tested. Optimum δ from simulation was applied to data from Holstein, Jersey and Brown Swiss dairy cattle to compare differences of adjusted and official genomic evaluations. From simulation, optimum δ was 0.4 for the heavy-tailed QTL distribution but only 0.1 or 0.2 for a normal distribution. The previous formula had slower response than unweighted selection in early generations and did not recover by generation 20. Long-term response was slightly greater with the new formulas than with unweighted selection; the linear formula may be best for routine use because of more progress in early generations compared to nonlinear formula. Official and adjusted U.S. evaluations based on actual genotypes and estimated marker effects were correlated by 0.994 for Holsteins and Jerseys and 0.989 for Brown Swiss using linear weighting of allele frequency, which was higher than nonlinear weighting. The difference between adjusted and official evaluations was highly correlated negatively with an animal's average genomic relationship to the population. Thus, strategies to reduce genomic inbreeding may achieve almost as much long-term progress as selection of favorable minor alleles.