If you say that, overall, the prevalence of cerebral palsy (CP) is perhaps 1 or 2 per 1000 in industrialized countries, as high as 4 per 1000 in some studies, and higher in select populations, few will argue with you. A statistical or epidemiological sort like me, however, might note that it depends on the definition of prevalence. Think what you may of such a person, the definition of prevalence is important, and many versions of it are employed in the literature. We cannot really know what the prevalence of CP is if it is uncertain what prevalence means. Whatever the definition, Peter Baxter1 recently drew our attention to the fact that the prevalence of CP has apparently not changed much over a long period of time. This may seem disappointing, but as Dr Baxter noted, real progress in preventing CP has been achieved. This progress has not resulted in overall reductions of prevalence, due in part to increases in survival of very preterm and very low birthweight infants. Prevalence of CP in these subgroups has fallen, but remains much higher than the prevalence overall.2 As researchers measure gains made in preventing CP and reducing its prevalence, it is natural to want to compare results across studies. In order for such comparisons of prevalence to be meaningful, a common definition would help. Failing that, the definitions used must be understood before any conclusions about differences in resulting prevalence estimates can be drawn. What then is prevalence? Prevalence is generally defined to be the number of cases (of CP for example) in a given place and time divided by the number of all persons in that place and time. Examine a number of studies, however, and you will find an almost equal number of definitions at work. Winter et al.3 employ a hybrid definition of birth prevalence; Moster et al.4 apparently use a kind of birth prevalence at age 4 and above; Kirby et al.5 use a traditional definition of prevalence at age 8; and Boyle et al.6 use a more or less traditional definition of prevalence based on a nationwide random sample. Among the cited studies, I would point to Winter et al.3 and Kirby et al.5 as good examples of how to present the methodology used to calculate prevalence. Sections in Winter et al. labeled ‘numerator data’ and ‘denominator data’ are particularly helpful; any study of prevalence should make it crystal clear how the numerator and denominator were determined. In Moster et al.4 readers are left to piece together the numerator and denominator by reading across multiple sections of the paper; I fear that most will not bother to do so. Do differences in definition really make a significant difference in resulting prevalence estimates? Certainly examples can be contrived wherein different definitions would result in dramatically different estimates of prevalence. Whether strict definitional differences have resulted in large differences in actual studies is less likely. Many other factors may be contributing to observed differences in prevalence estimates. For example, Boyle et al.6 exclude so-called ‘acquired’ cases of CP, meaning those cases resulting from events occurring more than 28 days after birth. Kirby et al.,5 on the other hand, include such cases. The effect of these different exclusion criteria may well outweigh that of other definitional differences. Nevertheless, more uniformity in definitions would undoubtedly increase the value of between-study comparisons, and improve the chances of detecting meaningful differences in the prevalence of CP. In the study by Andersen et al.7 we find further evidence of a decline in the prevalence of CP, this time among children born moderately preterm. The definition of prevalence used in this study is uniform across time periods and across a number of European registries of CP. That the definition does not match those used in Kirby et al.,5 Moster et al.,4 or other studies should be considered before any other comparisons are made between the prevalence estimates reported.