Cranial capacity is one of the most studied features used in human paleoneurology. According to the completeness of the fossil skulls, direct or indirect methods have been used in estimating the cranial capacity, including water displacement, seeds, and regression statistics. However, two problems always exist when dealing with fossil skulls: the incompleteness of the specimens and the small sample size. Both limits can introduce an important uncertainty when making inferences on cranial capacity through quantitative approaches. In a recent paper published in Quaternary International we improved traditional methods for estimating cranial capacity by using measurements on brain endocast instead of skull, and by using multivariate statistical method such as Principal Component Regression (PCR) and Partial Least Square Regression (PLSR) instead of simple linear models. In this study, modern human skulls and endocasts are used as training and test data. We then applied these methods on three Homo erectus specimens and one Late Pleistocene Homo sapiens individual. When compared with traditional ones, these methods show higher reliability, and the error of the estimations approaches 50 cc. This study stresses further the importance of methodological research and correlation analysis in paleoneurology.