The fragile X mental retardation syndrome is caused by unstable expansion of a CGG repeat. Two main types of mutation have been categorised. Clinical expression is associated with the presence of the full mutation, while subjects who carry only a premutation do not have mental retardation. Premutations have a high risk of transition to full mutation when transmitted by a female. We have used direct detection of the mutations to characterise large families who illustrate the wide variation in penetrance which has been observed in different sibships (a feature often called the Sherman paradox). A family originally found to show tight genetic linkage between the factor 9 gene and the fragile X locus was reanalysed, confirming the original genotype assignments and the observed linkage. The size of premutations was measured by Southern blotting and by using a PCR based test in 102 carrier mothers and this was correlated with the type of mutation found in their offspring. The risk of transition to full mutation was found to be very low for premutations with a size increase (DELTA) of about 100 bp, increasing up to 100% when the size of premutation was larger than about 200 bp, even after taking into account (at least partially) ascertainment bias. These results confirm and extend those reported by Fu et al (1991) and Yu et al (1992) and explain the Sherman paradox. The low risk of transition to full mutation of small premutations leads to the prediction that carriers of such alleles may be more frequent in the population than was previously expected for fragile X carriers, and we have indeed observed a premutation in a man with no a priori risk. Possible mechanisms that could account for the sex biased expansion of the CGG repeat are discussed in relation to the absence of such bias in expansion at the myotonic dystrophy locus.