Sampling rare religious populations in the United States poses a number of challenges. There are no official statistics on religious affiliation at a low-level, requiring pre-survey county-level estimates to be assembled from a variety of sources and estimated using small-area techniques. With less than 2% incidence for rare populations like U.S. Jews and Muslims, the cost of screening out ineligible households places great pressure on study budgets. To manage costs, counties are then grouped into strata based on estimated incidence for sample design and management. Sample is allocated to strata and landline and mobile frames using a nonlinear solver to optimise an objective function based on effective sample size subject to various constraints. Finally, weighting in the absence of official data on the population of interest to serve as post-stratification targets is a challenge, requiring the careful building up of post-stratification adjustments in an iterative fashion between adjustments for the whole of sample to general population targets and derived estimates for the population of interest. The techniques discussed are applicable to sampling other rare populations where official statistics are not available.