The use of whole-genome resequencing to obtain more information on genetic variation could produce a range of benefits for the dairy cattle industry, especially with regard to increasing milk production and improving milk composition. In this study, we sequenced the genomes of eight Holstein bulls from four half- or full-sib families, with high and low estimated breeding values (EBVs) of milk protein percentage and fat percentage at an average effective depth of 10×, using Illumina sequencing. Over 0.9 million nonredundant short insertions and deletions (indels) [1–49 base pairs (bp)] were obtained. Among them, 3,625 indels that were polymorphic between the high and low groups of bulls were revealed and subjected to further analysis. The vast majority (76.67%) of these indels were novel. Follow-up validation assays confirmed that most (70%) of the randomly selected indels represented true variations. The indels that were polymorphic between the two groups were annotated based on the cattle genome sequence assembly (UMD3.1.69); as a result, nearly 1,137 of them were found to be located within 767 annotated genes, only 5 (0.138%) of which were located in exons. Then, by integrated analysis of the 767 genes with known quantitative trait loci (QTL); significant single-nucleotide polymorphisms (SNPs) previously identified by genome-wide association studies (GWASs) to be associated with bovine milk protein and fat traits; and the well-known pathways involved in protein, fat synthesis, and metabolism, we identified a total of 11 promising candidate genes potentially affecting milk composition traits. These were FCGR2B, CENPE, RETSAT, ACSBG2, NFKB2, TBC1D1, NLK, MAP3K1, SLC30A2, ANGPT1 and UGDH. Our findings provide a basis for further study and reveal key genes for milk composition traits in dairy cattle.