Small area estimation of HIV incidence is challenging because sample surveys are designed to provide accurate estimates for “large areas” (e.g., national or regional) instead of “small areas” (i.e., subnational or subpopulation). In this presentation, we propose a generalized linear model in a Bayesian hierarchical framework to estimate small-area incidence as a parameter. For the model, we insert an informative prior based on an independent estimate of HIV incidence at large-area level, and use a likelihood based on incidence assay recency data. We first test our model on simulated data, and then apply it to the Malawi PHIA survey. Finally, we propose model evaluation techniques in this case where external cross-validation is impractical.