Engineered protein ligands have been successfully applied as molecular targeting agents in the clinic for diagnostic and therapeutic applications. Yeast surface display selections have shown effectiveness in discovery and evolution of ligands against a variety of target molecules to meet these ends. Often, these biomarkers are transmembrane proteins, which are made up with hydrophilic extracellular and intracellular domains and hydrophobic transmembrane domains. Due to these hydrophobic domains, full length transmembrane proteins are difficult to work with in aqueous systems. Thus, ligand selections are conventionally carried out against recombinant extracellular domains of these biomarkers. These molecules may be poor models of the corresponding full length, membrane associated proteins due to instability in the absence of the truncated domains, inadequacies in protein folding and post-translational modification, presence of non-natural tags required for protein purification and immobilization for selection, or denaturation during a variety of handling steps. Thus, ligand selection campaigns against these molecules often end with the evolution of ligands that bind the soluble target, but have diminished or abolished activity against the true biomarker of interest. The work presented here aims to provide a new toolkit for ligand selection that bypasses this translational hurdle by optimizing selections of yeast surface display libraries directly against target-expressing mammalian cell monolayers. Target specific binder enrichment is rigorously optimized and substantially improved through the use of a longer flexible linker in yeast surface display which enhances the recovery of even micromolar binding interactions that enable naïve library selections. Five ligand selection strategies using soluble target and cellular selection strategies are probed using a fibronectin domain and an affibody library, showing that methods partially or completely utilizing cellular selection strategies have a higher likelihood of isolating translatable binders while also developing new ligands that bind tumor vasculature targets CD276 and Thy1. Controlled valency reduction using the reducing agent dithiothreitol (DTT) to reduce yeast-displayed ligand levels to 3,000-6,000 ligands per cell enhances affinity discrimination between a 2 nM and 17 nM binder 16-fold, yielding a modular method for affinity-based cellular selection. This method is applied to discovery and affinity maturation of fibronectin domains targeting epithelial cell adhesion molecule (EpCAM) for oncological applications. Depletion of non-specific binding ligands is achieved using a pre-blocking strategy with disadhered mammalian cells, yielding a 14-fold selectivity advantage for a specific binder relative to a non-specific binder. Collectively, the results presented in this dissertation elucidate the factors that dictate cell-cell interactions within the context of yeast display ligand selections on mammalian cell targets and provide a suite of tools for ligand selections as well as a variety of targeting molecules with diagnostic and therapeutic applications.