Molecule-targeted therapies are being widely developed and deployed, but they are frequently less effective in clinical trials than predicted based upon preclinical studies. Frequently, only a single model or genetic background is utilized using diets that are not relevant to that consumed by most cancer patients, which may contribute to the lack of predictability of many preclinical therapeutic studies. Inhibition of epidermal growth factor receptor (EGFR) in colorectal cancer was used to investigate potential causes for low predictive values of many preclinical studies. The efficacy of the small molecule EGFR inhibitor AG1478 was evaluated using two mouse models, Apc(Min/+) and azoxymethane (AOM), both sexes on three genetic backgrounds, C57BL/6J (B6) and A/J (A) inbred strains and AB6F1 hybrids, and two diets, standard chow (STD) or Western-style diet (WD). AG1478 has significant anti-tumor activity in the B6-Apc(Min/+) model with STD but only moderately on the WD and in the AOM model on an A background with a WD but not STD. On the F1 hybrid background AG1478 is effective in the Apc(Min/+) model with either STD or WD, but has only moderate efficacy in the AOM model with either diet. Sex differences were also observed. Unexpectedly, the level of liver EGFR phosphorylation inhibition by AG1478 was not positively correlated with inhibition of tumor growth in the AOM model. Model-dependent interactions between genetic background and diet can dramatically impact preclinical results, and indicate that low predictive values of preclinical studies can be attributed to study designs that do not account for the heterogeneous patient population or the diets they consume. Better-designed preclinical studies should lead to more accurate predictions of therapeutic response in the clinic.