Monitoring provides the foundation for evaluating recovery of endangered species, yet many species lack monitoring programs designed to integrate a species’ unique attributes, specific monitoring objectives, and principles of statistical sampling theory. We developed a framework for monitoring and assessment of endangered light-footed Ridgway’s rails (Rallus obsoletus levipes) across their U.S. range, relative to multi-scale recovery goals. We created spatially explicit sample units and a sampling frame covering all potential habitat to facilitate range-wide probability sampling, and also built a model of the call-broadcast process commonly used to survey marsh birds that included heterogeneity in availability for detection and conditional detectability for each bird during each survey. We used the model to simulate 96 sampling strategies that included different levels of replication, multiple approaches for sample allocation amongst strata, and both simple random and weighted probability sampling (i.e., weights proportional to local rail abundance) of sample units within strata. Effective monitoring surveyed ≥20–30% of the sampling frame on ≥3 occasions, with weighted sample selection and more targeted sampling (50% of units) for strata that are key to species recovery. We also tested Bayesian N-mixture models for estimating abundance and show that multiple models provide reasonable estimates. This work lays the foundation for statistical sampling and multi-scale population estimation for an endangered bird, and for refinement of abundance estimation models. Moreover, this work provides a replicable process for building customized and statistically defensible sampling frameworks to assess recovery of endangered species that can used for other sensitive species.