Abstract: In multivariate statistical analyses, intact specimens are essentially required. In order to help researchers avoid having to discard many defective fossil specimens or greatly reduce the number of variables in their analyses, we developed a method based on the theory of linear regression for recovering missing data for defective fossil specimens. Using this method, missing value of measurements can be predicted based on other intact or defective equivalent specimens. Numerical tests have been carried out on the head and limb bones of extant Equus. The results show that our method, which is relatively insensitive to the quantity, preservational quality and type of available specimens, has satisfactory stability. The predictive accuracy is best for large specimens or measurements of large magnitude. Furthermore, our method is distinct from traditional linear regression methods in utilizing linear correlations between specimens rather than between variables, since the correlations are usually stronger in the case of the former procedure. Our method is simple in theory and practice, and should be broadly applicable to statistical analyses of fossil specimens, particularly if multivariate.