More recently, a study quite clearly summarized at Maju's blog has shown that there is more than one mutation that can produce blonde hair.
[T]he gene causing blond hair among Melanesians (and some relatives like Fijians) is not the same as those involved in blond hair in Europe. Mind you that it is not clear yet which are these European genes of blondism but it is clear that the Melanesian allele is not it[.]In the case of lactase persistance, there is a clear selective advantage to having the mutation, so it isn't surprising that multiple independent mutations that confer the phenotype would emerge.
Another less obvious example of what appears to be multiple, independent genetic causes for the same phenotype which apperas to have been driven by some sort of selective advantage is the set of mutations that lead to short stature in Pygmies and other Asian populations sometimes called Negritos. Rather than having deep origins the pre-date the Out of Africa era shared genetic histories of these populations, it appears that short stature is a relatively recent (i.e. Holocene era) adaptation to life in a marginalized hunter-gatherer population.
In the fairly simple examples mentioned above, only a handful of genes are at work. But, in the case of broad phenotypes like mental retardation, autism, or genetic predispositions to schizophrenia or bipolar disorder, the number of separate mutations that are expressed in ways that are currently indistinguishable in clinical diagnosis are probably in the hundreds or more each, with no single mutation playing a dominant role in any of these conditions at the level of specificity with which they are diagnosed today.
The Limits Of Current Genome Databases
Also instructive is the fact that no gene causing blond hair in Melanesians was present in any of the thousand genomes database individuals. The database can be very helpful in catching the lion's share of genetic variation for traits that are found a more than trivial frequency in large global populations, but rare traits in small populations can still easily be absent in the current set of published human genomes, even though each individual's genome provides hundreds of thousands of datapoints each.
Rich data on each individual in the sample may be useful for efforts to locate the ancestral origins of individuals in relation to each other with considerable precision despite having only a thousand or two individuals (and much smaller subsets for given geographic regions). But, they don't capture the long tail of genetic diversity in modern humans.