This paper evaluates village-level microfinance institutions in rural and semi-urban Thailand using data from a large stratified; clustered cross-sectional survey. As is typically the case, micro-finance institutions have emerged without the benefit of experimental controls. Our data therefore has neither randomized participation/ program placement; a pre-specified, narrow set of institutions with supposedly known operational and selection policies; nor reliable pre-survey information on the presence or absence of a various - possibly endogenously emerging - institutions. Despite the obvious selection problems, there is one great advantage: the variation in policies and institutions observed in the survey allow an assessment of policies correlated with success. This is extremely important to policymakers contemplating expansion of existing programs or new programs. Our data and analysis allow us to compensate somewhat for the lack of randomized controls. A key ingredient is the use of data from multiple instruments - questionairres and accounts from institutions themselves, surveys of member and non-member households and interviews with headmen to elicit village histories. Thus, the methods here would be applicable in other countries requiring only an investment in survey information. These various instruments allow us to highlight important institutional characteristics that are cor- related with proximate measures of institutional success: growth in membership, savings, and credit services. Examples of findings uncovered: the positive (negative) correlation of cash (rice, buffalo) loans with lending growth; the positive (negative) correlation of pledged (traditional, time deposit) savings accounts with saving growth, and the positive correlation of the provision of agricultural training and non-agricultural consultation with lending and savings growth, respectively. We also present evidence that villages with institutions, and villages with successful institutions in particular, differ from villages without institutions, or with unsuccessful institutions, in ways not related to the institutions themselves. That is, villages where institutions exist prior to, at the time of, or after a retrospective date tend to have more financially active households, but the villages are poorer, more agricultural, and more informal. Villages with successful lending institutions average significantly lower incomes and educational levels, and have less and tighter credit. Villages with successful savings institutions tend to have higher levels of credit, but more limited savings. Evidence suggests these characteristics predate the instititutions themselves. In short, there is ample evidence that the apparent impact of an institution on village members or the impact of a successful institution has much to do with the prior characteristics of the respective villages and members. Thus, an assessment of the direct impact of micro-finance institutions on household and household businesses must take this individual and village selection into account. Linear regressions and simple probits of the frequency of household business start-ups, credit constraints in business or agriculture, oc- cupational mobility, reliance on moneylenders, and asset growth would show that village-level institutions have insignificant or even perverse effects. Controlling for village-level selection and explicitly allowing for individual selection (correlated errors/ simultaneity in the membership and impact equations) eliminates the apparent perverse effects and estimates significant positive effects. In particular, institutions do seem to promote asset growth, reduce credit constraints in agriculture, reduce reliance on moneylenders and increase occupational mobility. Moreover, the institutional policies and characteristics correlated with institutional growth are linked to these direct impacts.