Milk composition traits exhibit a complex genetic architecture with a small number of major quantitative trait loci (QTL) explaining a large fraction of the genetic variation and numerous QTL with minor effects. In order to identify QTL for milk fat percentage (FP) in the German Holstein-Friesian (HF) population, a genome-wide association study (GWAS) was performed. The study population consisted of 2327 progeny-tested bulls. Genotypes were available for 44,280 SNPs. Phenotypes in the form of estimated breeding values (EBVs) for FP were used as highly heritable traits. A variance components-based approach was used to account for population stratification. The GWAS identified four major QTL regions explaining 46.18% of the FP EBV variance. Besides two previously known FP QTL on BTA14 (P = 8.91×10-(198)) and BTA20 (P = 7.03×10(-12)) within DGAT1 and GHR, respectively, we uncovered two additional QTL regions on BTA5 (P = 2.00×10(-13)) and BTA27 (P = 9.83×10(-5)) encompassing EPS8 and GPAT4, respectively. EPS8 and GPAT4 are involved in lipid metabolism in mammals. Re-sequencing of EPS8 and GPAT4 revealed 50 polymorphisms. Genotypes for five of them were inferred for the entire study population. Two polymorphisms affecting potential transcription factor binding sites of EPS8 (P = 1.40×10(-12)) and GPAT4 (P = 5.18×10(-5)), respectively, were highly significantly associated with the FP EBV. Our results provide evidence that alteration of regulatory sites is an important aspect of genetic variation of complex traits in cattle.