Data comprised information on herd life from two different sources: actual lifespan and type data. The optimum way of predicting breeding values (BVs) for herd life from both these sources is a multivariate best linear unbiased predictor (BLUP) analysis of lifespan and the linear type traits most closely related to it. To reduce computing requirements, we suggest a bivariate BLUP analysis, where the direct information is lifespan, measured in lactations and the indirect information is a phenotypic index of type traits weighted by their economic values. Such an index is shown to be almost as efficient as using the individual traits. Genetic correlations between functional lifespan and the linear type traits were estimated, and based on these foot angle (r = 0.22) udder depth (r = 0.14) and teat length (r = -0.44) were chosen to predict herd life. Breeding values for herd life were predicted in a BLUP analysis for around 500 000 heifers. The traits were lifespan and the phenotype index of foot angle, udder depth and teat length. For bulls with a minimum of 10 daughters BVs ranged from -0.9 lactations to +1.0 lactations, indicating a difference of approximately one lactation between the daughters of extreme bulls. For sires with at least 50 daughter records for lifespan, type information contributed little to the accuracy of herd life BVs.