Objective: To observe the longitudinal change of cognitive function of the elderly in Beijing. Method: A sample of people aged 55 and over in community were tested with MMSE at baseline and followed up five years later. Results: 241 out of 2047 were cognitive impaired at baseline survey. Of them there were 42.9% dead in following-up, with a death rate much higher than 19.8% of the cognitive intact persons. The difference in age and morbidity does not account for the disparity in mortality between cognitive impaired and non-impaired subjects. Among 1337 people who had finished MMSE test twice, the mean score of MMSE in following-up was 23.36, with 0.41 lower than the original mean score. Individually, 51.9% of them remained unchanged in MMSE score, 24.5% decreased and 23.6% obtained a higher score five years later. Conclusion: The decline of cognitive function in community elderly people is quite slow. Brief cognitive test as MMSE is to show further cognitive impairment, death in the next five years, as well as to identify the risk of the subjects with suspected dementia.