The units of analysis in this research are 3,363 persons of registration in or residence in 643 households of five big cities in Taiwan. The main aim is to explore the effects of demographic variables on accuracy of migration registration. Demographic variables are the relationship with head of household, sex, age, marital status, education, and occupation. The accuracy of migration registration are indicated by yes/no move-in registration, yes/no registration without residence in, yes/no residence in without registration. Using discriminant analysis, we try to find out the demographic variables which can discriminate the accuracy of migration registration the most significantly. Then we use log-linear model to estimate the frequencies of yes/no accurate migration registration and their odds with three most significant discriminant variables. We found that the most significant discriminant variables discriminating the accuracy of migration registration for persons under age 15 are the relationship with head of household and age. Persons who are not members of nuclear family of the head of household or age under 6 have higher probability of inaccuracy of migration registration. For those persons age 15 and above, the relationship with head of household and education can discriminate yes/no mov-in registration and yes/no residence in without registration. Those persons who are not the members of nuclear family of head of household and college graduates have higher probability not registering for their residence. However the most significant discriminant variables to discriminate yes/no registration in without residence in this household is marital status; the single has 10 times of probability of registration inaccuracy as the married. Finally, this research suggests that all the persons with demographic characteristics which have higher probability of inaccuracy of migration registration need to be checked more often.