Winter hardiness is a major constraint for autumn sown crops in temperate regions, and thus an important breeding goal in the development of new winter wheat varieties. Winter hardiness is though influenced by many environmental factors rendering phenotypic selection under field conditions a difficult task due to the irregular occurrence of winter damage that is alternated by complete winter kills of all selection candidates tested in field trials. Controlled frost tolerance tests in growth chamber experiments are on the other hand even with few genotypes often costly and laborious, which makes a genomic breeding strategy for early generation selection an interesting alternative. The aims of this study were thus 1) to compare the merit of marker-assisted selection using the major frost tolerance QTL Fr-A2 with genomic prediction for winter hardiness and frost tolerance, and 2) to assess the potential of combining both measures with a genomic selection index using a high density marker map or a reduced set of pre-selected markers. A training population including 170 lines with winter hardiness data from 2012, another training population of 200 lines from a controlled frost test experiment in 2017, and an independent validation population of 130 lines that was assessed for winter hardiness in the field 2018 were for this purpose subject to an integrated analysis. Cross-validation within the training population for frost tolerance showed the merit of marker-assisted selection for the copy number variation at Fr-A2 especially when its effect was upweighted in genomic prediction models, while a new QTL on chromosome 3A could be identified with the winter hardiness data from 2012. Combining both measures in a genomic selection index increased the accuracy of genomic predictions for the independent validation population 2018 in comparison to training with winter hardiness data alone. The prediction accuracy for winter hardiness could moreover be maintained in this independent validation population when using pre-selected sets of 20-40 instead of several thousand markers, which is of high relevance when employing cost reducing fingerprinting techniques such as targeted genotyping-by-sequencing in practical breeding programs. Genomic selection showed thus large potential to improve or maintain the performance of winter wheat for this difficult, costly and laborious to phenotype trait.