Diffusion-weighted imaging (DWI) is commonly used to assess irreversibly infarcted tissue but its accuracy is challenged by reports of diffusion lesion reversal (DLR). We investigated the frequency and implications for mismatch classification of DLR using imaging from the EPITHET (Echoplanar Imaging Thrombolytic Evaluation Trial) and DEFUSE (Diffusion and Perfusion Imaging Evaluation for Understanding Stroke Evolution) studies. In 119 patients (83 treated with IV tissue plasminogen activator), follow-up images were coregistered to acute diffusion images and the lesions manually outlined to their maximal visual extent in diffusion space. Diffusion lesion reversal was defined as voxels of acute diffusion lesion that corresponded to normal brain at follow-up (i.e., final infarct, leukoaraiosis, and cerebrospinal fluid (CSF) voxels were excluded from consideration). The appearance of DLR was visually checked for artifacts, the volume calculated, and the impact of adjusting baseline diffusion lesion volume for DLR volume on perfusion-diffusion mismatch analyzed. Median DLR volume reduced from 4.4 to 1.5 mL after excluding CSF/leukoaraiosis. Visual inspection verified 8/119 (6.7%) with true DLR, median volume 2.33 mL. Subtracting DLR from acute diffusion volume altered perfusion-diffusion mismatch (T(max)>6 seconds, ratio>1.2) in 3/119 (2.5%) patients. Diffusion lesion reversal between baseline and 3 to 6 hours DWI was also uncommon (7/65, 11%) and often transient. Clinically relevant DLR is uncommon and rarely alters perfusion-diffusion mismatch. The acute diffusion lesion is generally a reliable signature of the infarct core.