Our study is inherently limited by its secondary, exploratory nature, cross-sectional design, and small sample size. We used retrospective self-report of anemia diagnosis coupled to current residential drinking water arsenic levels; we are therefore unable to establish a temporal sequence in which exposure preceded outcome. Consequently, our results can be interpreted only as hypothesis generating. Our use of self-reported anemia diagnoses and exclusion of women not diagnosed while residing at the study address are also likely to have misclassified some women, and would have benefitted from clinical confirmation, which was unfortunately unavailable. However, we have no reason to anticipate that anemia misclassification would have varied by drinking water arsenic exposure and thus any bias is likely to have been towards the null hypothesis. Further, we found similar results in sensitivity analyses restricted to cases diagnosed for the study pregnancy. The limited number of study participants decreased the statistical power and led to imprecise effect estimates. However, our post-hoc power analysis suggested that a modest increase in sample size to n=246 will provide statistical power sufficient to detect an association between pregnancy anemia and drinking water arsenic exposure.