Because foods are consumed in combination, it is difficult in observational studies to separate the effects of single foods on the development of diseases. A possible way to examine the combined effect of food intakes is to derive dietary patterns by using appropriate statistical methods. The objective of this study was to apply a new statistical method, reduced rank regression (RRR), that is more flexible and powerful than the classic principal component analysis. RRR can be used efficiently in nutritional epidemiology by choosing disease-specific response variables and determining combinations of food intake that explain as much response variation as possible. The authors applied RRR to extract dietary patterns from 49 food groups, specifying four diabetes-related nutrients and nutrient ratios as responses. Data were derived from a nested German case-control study within the European Prospective Investigation into Cancer and Nutrition-Potsdam study consisting of 193 cases with incident type 2 diabetes identified until 2001 and 385 controls. The four factors extracted by RRR explained 93.1% of response variation, whereas the first four factors obtained by principal component analysis accounted for only 41.9%. In contrast to principal component analysis and other methods, the new RRR method extracted a significant risk factor for diabetes.