Data from 321 404 finishing bulls of the Bavarian progeny field test, based on slaughterhouse information, were analysed to further develop breeding value estimation. Because of a small number of observations per farm, herd-year-season clusters were used in the linear model. Models with fixed and random contemporary group definitions were compared regarding variance components and breeding values. Estimates for the heritabilities of marker value and net gain were 0.11 and 0.16, respectively, when contemporary group was defined as a fixed effect and 0.11 and 0.12 when defined as uncorrelated random. As several local organizations are involved in running the breeding programme for Simmental in Bavaria, unproven sires will have progeny only within a certain region. With an increasing number of observations per cluster, as observed in recent years, the Pearson correlation coefficient and the rank correlation of breeding values from both models were higher than 0.90. Therefore a fixed model is recommended. Although the heritabilities are low, a large number of effective progeny per sire yields a sufficient accuracy of estimated breeding values. The progeny test in the field and the performance test on station are the most important tools in Bavarian AI sire evaluation of beef traits. The additional implementation of the field test delivers more accurate breeding values at low costs and does not prolong the generation interval, because the information on beef traits is available before the female progeny finish their first lactation.