One of the firmly held concepts in human molecular genetics has been that, if we can understand the details of specific genetic mutations and their effects on protein products, we will be better able to correlate genotype with phenotype. One of the promises of this concept is that such a knowledge base will move clinical genetics into a predictive mode: knowledge of the mutant alleles responsible for a disease would permit an accurate prediction of the prognosis and a better-informed selection among therapeutic strategies for any individual patient. As the mutations have been identified for a series of diseases, it has become clear that the correlation between genotype and phenotype is often incomplete. What has emerged is the recognition that, for many diseases, only a subset of all mutations reliably predicts phenotype. We propose that there are two thresholds relating mutant protein function to phenotype: a level below which the severe phenotype will always be observed and another level above which the phenotype will be uniformly mild (fig. 1). Between these two thresholds is an indeterminate range, in which mutations would not correlate with phenotype and additional unlinked genes and/or environmental factors would influence the final phenotype. For many diseases, both thresholds may be observed; but, for others, one or both thresholds may not be seen. For example, there may not be any mutation with sufficiently low protein function to result uniformly in the severe phenotype and/or with sufficiently adequate function to be associated consistently with the mildest phenotype.