The American Community Survey (ACS) sampling procedure for group quarters facilities is designed to obtain state-level estimates. However, estimates for the total resident population – household and group quarters residents combined – are published at the census tract and county levels as well. A consequence of the sampling procedure is that often little to no group quarters data are collected for small areas. To improve estimates at lower levels of geography, four imputation procedures are proposed. The objective is to obtain, through sampling or imputation, county by major type group representation for 1- and 3-year estimates and tract by major type group representation for 5-year estimates. Entire person records are imputed, using the sampled group quarters records, into areas with little sample coverage. Donors are chosen based either on geographic proximity or the similarity of the surrounding housing population. To evaluate each imputation procedure, a group quarters population is simulated using Census 2000 short-form data. From this population, twenty-five independent ACS samples are drawn and each imputation procedure is applied to the simulated samples. The resulting 1-, 3-, and 5-year imputation-based estimates of basic demographic variables are compared to the design- based estimates and to the simulated population values. The bias of the imputation-based estimates is found to be larger than the design-based estimates, but the variance is smaller.