Background: Up to 20% of lacunar infarcts are misdiagnosed as cortical infarcts clinically and vice versa. The reasons for this discrepancy are unclear. We assessed clinical and imaging features which might explain this ‘clinical-imaging dissociation’. Methods: Patients with an acute stroke syndrome (cortical or lacunar) underwent magnetic resonance imaging including diffusion-weighted imaging (DWI). We recorded DWI-positive infarcts, proximity to cortex for small subcortical infarcts. We examined factors associated with clinical-imaging dissociation. Results: 137 patients with a mild cortical or lacunar syndrome had an acute ischaemic lesion on DWI. Of these, 21/93 (23%) with a cortical syndrome had an acute lacunar infarct and 7/44 (16%) with a lacunar syndrome had an acute cortical infarct. From 72 patients with an acute lacunar infarct on DWI, lesion proximity to cortex (odds ratio (OR) 14.5, 95% confidence interval (CI) 1.61 to 130.1), left hemisphere location (OR 8.95, 95% CI 1.23 to 64.99) and diabetes (OR 17.1, 95% CI 1.49 to 196.16) predicted clinical-imaging dissociation. On multivariate analysis of all 137 patients, clinical-imaging dissociation was associated with diabetes (OR 7.12, 95% CI 1.86 to 27.2). Conclusions: Clinical-imaging dissociation occurs in a fifth of patients with mild stroke. Lacunar infarcts lying close to cortex are more likely to cause cortical symptoms. Diabetes is associated with any clinical-imaging mismatch. Stroke misclassification which can arise with clinical classification alone should be minimised in research by verification with high sensitivity imaging.