A salient feature of mammalian X dosage compensation is that X-inactivation occurs without regard to the parental origin of either active or inactive X. However, there are variations on the theme of random inactivation, namely paternal X inactivation in marsupials and in placental tissues of some mammals. Whether inactivation is random or paternal seems to depend on the time when this developmental program is initiated. As deletions of the X inactivation center (XIC/Xic) and/or the X inactive specific transcript (XIST/Xist) gene result in failure of cis X-inactivation, mutations in genes from this region might lead to preferential inactivation of one X chromosome or the other. The Xce locus in the murine Xic is considered a prototype for this model. Recent studies suggest that choice involves maintaining the activity of one X, while the other(s) by default is programmed to become inactive. Also, choice resides within the XIC, so that mutations elsewhere, although perhaps able to interfere with cis inactivation, are not likely to affect the X chromosome from only one parent. Mutations affecting the choice of active X will be more difficult to detect in humans than in inbred laboratory mice because of the greater allelic differences between maternal and paternal X chromosomes; some of these differences predispose to growth competition between the mosaic cell populations. I suggest that the skewing of inactivation patterns observed in human females most often occurs after random X inactivation, and is due mainly to cell selection favoring alleles that provide a relative growth advantage.