Considerable progress has been made in the development of approaches for HIV incidence estimation based on a cross-sectional survey for biomarkers of recent infection. Multiple biomarkers when used in combination can increase the precision of cross-sectional HIV incidence estimates. Multi-assay algorithms (MAAs) for cross-sectional HIV incidence estimation are hierarchical stepwise algorithms for testing the biological samples with multiple biomarkers. The objective of this paper is to consider some of the statistical challenges for addressing the problem of missing biomarkers in such testing algorithms. We consider several methods for handling missing biomarkers for (1) estimating the mean window period, and (2) estimating HIV incidence from a cross sectional survey once the mean window period has been determined. We develop a conditional estimation approach for addressing the missing data challenges and compare that method with two naïve approaches. Using MAAs developed for HIV subtype B, we evaluate the methods by simulation. We show that the two naïve estimation methods lead to biased results in most of the missing data scenarios considered. The proposed conditional approach protects against bias in all of the scenarios.