Since the identification of mutations in MECP2 in girls and women with apparent Rett syndrome, numerous efforts have been made to develop phenotype‐genotype correlations. These studies have produced conflicting results in part related to use of different clinical severity scales, different diagnostic criteria, and different stratification by age and mutation group as well as the possible effects of unbalanced X‐chromosome inactivation. The present study applied a revised ordinal scoring system that allowed for correction for patient ages. We analyzed 85 patients with mutation in MECP2. Sixty‐five (76%) had one of eight common mutations. Patients with missense mutations had lower total severity scores and better language performance than those with nonsense mutations. No difference was noted between severity scores for mutations in the methyl‐binding domain (MBD) and the transcriptional repression domain (TRD). However, patients with missense mutations in TRD had the best overall scores and better preservation of head growth and language skills. Analysis of specific mutation groups demonstrated a striking difference for patients with the R306C mutation including better overall score, later regression, and better language with less motoric impairment. Indeed, these patients as a group accounted for the differences in overall scores between the missense and nonsense groups. Thus, the impact of specific mutations coupled with possible variation in X‐chromosome inactivation must be considered carefully in the derivation of phenotype‐genotype correlations. These results emphasize the limitations of such analyses in larger mutation groups, either by type or position. © 2003 Wiley‐Liss, Inc.