<i>Background:</i> Cognitive impairment is an important but underrecognised consequence of stroke. We investigated whether a subset of items from the NIH Stroke Scale (NIHSS) could yield valid information on cognitive status in a group of stroke patients. <i>Methods:</i> 149 stroke patients from the Göteborg 70+ Stroke Study were investigated after 18 months. We extracted 4 items corresponding to the NIHSS items on orientation, executive function, language and inattention. Scores on this subset of 4 NIHSS items (Cog-4) and the Mini-Mental State Examination (MMSE) were evaluated against a reference diagnosis of severe cognitive impairment. <i>Results:</i> The area under the receiver-operator curve (AUC) plotted for the Cog-4 scale against the diagnosis of severe cognitive impairment was 0.78; the MMSE had a slightly better diagnostic precision, with an AUC of 0.84. Making the executive task more difficult increased the precision of the Cog-4, raising the AUC to 0.81. <i>Conclusions:</i> A composite score based on 4 NIHSS items is almost as good as the MMSE in detecting severe cognitive impairment. Ideally, dedicated measures of cognition should be employed as a matter of course after stroke, but in their absence, the Cog-4 subscale provides an indication of cognitive functioning.