Background: The National Institute of Health Stroke Scale (NIHSS) results at baseline and after 1 year have never before been accounted for within an unselected population-based stroke sample. Neither has it been shown which individual items in the scale are the most important ones for the outcome in terms of death or dependency after 1 year. Methods: The subjects were all patients within a municipality who had their first-ever non-subarachnoidal stroke during 1 year (n = 377). Impairment was evaluated at baseline (within 24–48 h) and after 1 year with the 15-item version of the NIHSS. At the 1-year follow-up, the Modified Rankin Scale was used in order to determine which patients were dependent. Predictors of death and dependency were analysed in logistic regression models. The different NIHSS items, age and gender were used as independent variables. Results: The median NIHSS score was 6 (interquartile range 3–12) at baseline and 1 (interquartile range 0–3) at the 1-year follow-up, when 33% of the patients had died. Of patients scoring less than 4 on baseline NIHSS, 75% were functionally independent after 1 year. Seventeen per cent were functionally dependent and 8% were dead. Independent predictors of death were: age, questions, commands, gaze, alertness and sensation. Independent predictors of dependency were: age, commands, alertness and motor leg. Conclusions: Baseline NIHSS predicts the outcome after 1 year at the group level. Age and any reduction of the level of consciousness on arrival were associated with bad outcome after 1 year.