The importance of polymorphisms in the dihydropyrimidine dehydrogenase (DPD) gene (DPYD) for the prediction of severe toxicity in 5-fluorouracil (5-FU) based chemotherapy has been controversially debated. As a key enzyme in the catabolism of 5-FU, DPD is the top candidate for pharmacogenetic studies on 5-FU toxicity, since a reduced DPD activity is thought to result in an increased half-life of the drug, and thus, an increased risk of toxicity. Here, we review the current knowledge on well-known and frequently studied DPYD variants such as the c.1905+1G>A splice site variant, as well as the recent discoveries of important functional variation in the noncoding regions of DPYD. We also outline future directions that are needed to further improve the risk assessment of 5-FU toxicity, in particular with respect to metabolic profiling and in the context of different combination therapeutic regimens, in which 5-FU is used today.