The Amish marry within their faith. Although they accept converts, there are very few, so there is almost no inward gene flow. They descend almost entirely from about 200 18th century founders. On the other hand, there is considerable outward gene flow, since a significant fraction of Amish youth do not choose to adopt the Amish way of life. In recent years, something like 10-15% of young Amish leave the community In the past, the defection rate seems to have been higher, more like 18-24%. Defection is up to the individual - there are no exterior barriers against Amish who want to participate in modern society.
Since the Amish have very high birth rates ( > 6 children per family), their numbers have increased very rapidly, even though there is a substantial defection rate. There were about 5,000 descendants of the original 200 by 1920, and today  there are about 280,000 Amish.
Every way of life selects for something, but the Amish way of life is so different that natural selection in that population should be noticeably different from that in the general US population. It seems likely that the Amish have undergone selection for two specific traits, due to their unusual social and reproductive pattern.
First, they were almost certainly selected for higher fertility. A recent study (Milot et al., 2011) found evidence of this kind of selection in preindustrial French Canadians, who, like the Amish, went through a very rapid population expansion.
Second, and more interesting, the Amish have probably experienced selection for increased Amishness - an increase in the degree to which Amish find their lifestyle congenial, since those who like it least, leave. We have called this kind of differential emigration 'boiling off'. Obviously, if some of the soup boils off, what is left is more concentrated.
This boiling off is essentially truncation selection. If we assume a normal distribution, the loss of the least plain 10% corresponds to the loss of everyone more than 1.25 standard deviations below the Amish mean. If we assume a narrow-sense heritability of 0.3 and use a scale similar to that for IQ, the Amish gain about 1 point of plainness per generation. Not counting possible selection for this kind of personality in Europe, before they settled North America, the Amish have spent about ten generations under this kind of selection. Therefore their 'plainness', their Amish quotient (AQ), might have increased by about 0.6 standard deviation. During most of the period for which we have sufficient in- formation, the defection rate was significantly higher than 10%, so this may be a conservative estimate. Although there are certainly other factors that might influence the defection rate, such as increasing differences between the Amish way of life and that of their neighbors, increasing plainness would tend to reduce the defection rate over time.
The Amish have some genetic problems because of genetic drift in a small population, and those have received a fair amount of attention from medical geneticists. However, in our opinion, their social pattern probably drives strong selection for a particular flavor of personality, which is downright fascinating and worthy of further investigation. One could, with difficulty and a lot of investment, identify dimensions of a hypothetical AQ. It would likely include affinity for work, perseverance, low status competition, respect for authority, conscientiousness, community orientation, and so on. We proposed (Cochran, Hardy, & Harpending, 2006) a similar mechanism to account for Ashkenazi Jewish evolution in Medieval times selecting for ability and success in white collar occupations.