Based on the interaction concept of ICF theoretical framework，this study constructs a disability measurement index system from a multi-dimensional perspective．First of all，we get the total score of disability in multi-dimension by assignment method． According to the score，it was divided into five functional states: severe disability partial disability，moderate disability，mild disability and health．At the same time，we construct death probability model and disability state transition probability model，and the results show that the disability classification standard is reliable．Secondly，we use the tracking survey data of 2011 and 2014 in CLHLS database to empirically measure the transition probability of multi state disability by constructing the multi-state transition probability matrix. With the data of the sixth national census，we estimate the scale of disabled population aged 65 and above，and the maintenance time of each state and life expectancy．The results show that the three-year transfer probability of the initial healthy elderly is the highest, and the mortality rate is the lowest; while the mortality rate of the elderly with severe disability is the highest, and the probability of maintaining severe disability and transferring to the disabled state of overweight degree is also the highest．In addition，the conclusion of this study shows that the estimation of the disabled state transition probability based on the data is more accurate than the modeling estimation; the failure scale and life expectancy based on the multi state transition probability matrix evaluation are more reliable than the existing hierarchical measurement results．
