Objective To explore the association between underweight and mortality from all cause Method We examined the relationship of underweight and mortality among subjects who participated in the survey of cardiovascular risk factors (in 1984-1986 and in 1992). Relative Risks (RRs) were estimate from the COX propor tional hazard model by using a BMI between 18.5-24.9kg/m2 as reference c ategory Results During an average 12.40 years followup, we documented 307 deaths of which 158 were from cancer, 93 from cardiovascular diseases, and 56 from other causes W e observed a Lshaped relation between BMI and overall mortality.(P for trend< 0.01); In multivariated analyses, we found that the RR of mortality for a BMI o f <18.5kg/m2 was 1.58 (95% CI 1.16-2.14) as compared with the risk a mong subjects with BMI of 18.5-24.9kg/m2 The RR of overall mortality among s ubjects with BMI of <18.5kg/m2 was still significant after the exclusion of subjects who died from cancers during the first five years follow-up Conclusion Total mortality was increased among underweight people, and this association can not been completely explained by smoking and early deaths from caner.