This study examined cross-validation and test-retest reliability of questions and questionnaire indices commonly used for twin zygosity classification. Mothers of 58 monozygotic (MZ) and 52 dizygotic (DZ) same sex twin pairs were interviewed by telephone to answer questions regarding the similarity of their twins (mean age = 14.6 ± 2.8 years). A logistic regression equation correctly classified 91% of both MZ and DZ twin pairs in our sample using 7 of the 12 zygosity questions. The internal consistency for the total questionnaire (Cronbach's alpha) was 0.88. The median two month temporal stability estimate for the individual questions was r = .56 and r = .79 for the test total. For the cross-validation, zygosity classification indices taken from 9 previous studies were applied to our sample and compared to classification according to DNA microsatellite analyses (agreement range = 44 to 100%). The accuracy of the classification indices was significantly lower than the original studies for 62% of the comparisons. If zygosity determination with DNA markers or blood group typing for all subjects is not feasible, rather than using classification indices based on other studies, an optimal classification scheme can be achieved by using a zygosity questionnaire of which the reliability and validity of the questions is established in a random subsample of the same twin cohort.