The tenth human fibronectin type three domain ((10)Fn3) is a small (10 kDa), extremely stable and soluble protein with an immunoglobulin-like fold, but without cysteine residues. Selections from (10)Fn3-based libraries of proteins with randomized loops have yielded high-affinity, target-specific antibody mimics. However, little is known about the biophysical properties of such antibody mimics, which will determine their suitability for in vitro and medical applications. We characterized target binding and biophysical properties of two related (10)Fn3-based antibody mimics that bind vascular endothelial growth factor receptor two (VEGF-R2). The first antibody mimic, which has a dissociation constant (K(d)) of 13 nM, is highly stable [melting temperature (T(m))=62 degrees C] and soluble, whereas the second, which binds VEGF-R2 with 40 x higher affinity, is less stable (T(m) < 40 degrees C) and relatively insoluble. We used our understanding of these two (10)Fn3 derivatives and of wild-type (10)Fn3 structure to engineer the next generation of antibody mimics, which have an improved combination of high affinity (K(d)=0.59 nM), stability (T(m)=53 degrees C) and solubility. Our findings illustrate that (10)Fn3-based antibody mimics can be engineered for favorable biophysical properties even when 20% of the wild-type (10)Fn3 sequence is mutated in order to satisfy target-binding requirements.