In our study “Anesthesia provider model, hospital resources, and maternal outcomes” (Needleman and Minnick 2009), we examined the relationship between models for delivery of anesthesia services in obstetrics and maternal outcomes, notably maternal death, anesthesia-related complications, and other complications. The relationship of anesthesia provider type and adverse outcomes has been the focus of several studies. Debates about reimbursement rules and, at the state level, debates about certified registered nurse-anesthetist (CRNA) scope of practice and supervision have been ongoing. We return to the discussion of our findings within this policy context at the end of this paper. Our study used data from 365 hospitals in six states that responded to a survey on the organization and staffing of obstetrical and obstetrical anesthesia services and 1.14 million obstetrical discharges from those hospitals over a 3-year period. Using logistic regression and a variety of propensity models to control for selection, we found no evidence that maternal mortality or anesthesia and other complication rates were higher in hospitals that relied on CRNAs only or both anesthesiologists and CRNAs for obstetrical anesthesia, compared with hospitals using only anesthesiologists for these services. All comparisons of mortality or anesthesia-related complications and provider models found either small, nonsignificant differences or small statistically significant lower rates of complications in hospitals using CRNAs. We then suggested that our findings may have potential importance for policy in that, at least in the area of obstetrical services, there may be no gain in anesthesia safety from restricting which licensed providers can provide these services. Use of CRNAs makes it possible to provide obstetrical anesthesia coverage where anesthesiologists are not available. We suggested other paths for improving safety beyond restrictions based on training. In their commentary on our study, Neuman, Schwartz, and Fleisher (2010), while recognizing that we controlled for a range of potential confounders, argued that our study may have been limited by unobserved differences in patient risk. They offered examples of potential sources of bias not measured and, based on their speculation of how these could impact the results, urged caution in using our findings to guide specific health policy decisions. Expanding beyond our study's results and implications, Neuman and his colleagues raise issues as to how policy makers, physicians, and patients should use findings from any observational study in their decision making. In this response, we first address the general approach to dealing with the limitations to observational studies. Then we discuss the comments as they apply to our work and offer some new analyses to address some of Neuman and colleagues' sources of bias. We conclude with our perspective on the broader issues about when and whether to draw inferences for action from observational studies.