Traumatic brain injury (TBI) has a complex and heterogeneous pathology. It is frequently difficult to predict the neurological deterioration of patients with TBI, and unpredictable change may occur even when TBI is mild to moderate. When computed tomography (CT) findings are considered to be inconsistent with the traumatic origin or with the neurological deterioration of patients observed on admission, magnetic resonance imaging (MRI) is employed based on the standards of our ethical committee. In this retrospective study, we compared CT and diffusion weighted imaging (DWI) of patients with mild to moderate TBI in the very acute phase. When the high-intensity lesions on DWI are larger than the high-density lesions on CT images, we defined the imaging finding as a 'CT-DWI mismatch'. Between January 2010 and December 2013, 92 patients were inspected using both CT and MRI at admission, and we detected a CT-DWI mismatch in 35 patients. CT-DWI mismatch was 92.6% (95% confidence interval 79.8-97.9) sensitive and 84.6% (95% confidence interval 79.3-86.3) specific for the prediction of enlargement of the hemorrhagic lesions on repeat CT. CT-DWI mismatch is considered to be useful as one of the predictors of the enlargement of hemorrhagic lesions in patients with mild to moderate TBI.