Genetic growth curves of boars in a test station were predicted on daily weight records collected by automated weighing scales. The data contained 121 865 observations from 1477 Norwegian Landrace boars and 108 589 observations from 1300 Norwegian Duroc boars. Random regression models using Legendre polynomials up to second order for weight at different ages were compared for best predicting ability and Bayesian information criterion (BIC) for both breeds. The model with second-order polynomials had best predictive ability and BIC. The heritability for weight, based on this model, was found to vary along the growth trajectory between 0.32–0.35 for Duroc and 0.17–0.25 for Landrace. By varying test length possibility to use shorter test time and pre-selection was tested. Test length was varied and compared with average termination at 100 kg, termination of the test at 90 kg gives, e.g. ∼2% reduction in accuracy of estimated breeding values (EBV) for both breeds and termination at 80 kg gives ∼5% reduction in accuracy of EBVs for Landrace and ∼3% for Duroc. A shorter test period can decrease test costs per boar, but also gives possibilities to increase selection intensity as there will be room for testing more boars.