Transmembrane proteins constitute a large fraction of cellular proteins, and specific interactions involving membrane-spanning protein segments play an important role in protein oligomerization, folding, and function. We previously isolated an artificial, dimeric, 44-amino acid transmembrane protein that activates the human erythropoietin receptor (hEPOR) in trans. This artificial protein supports limited erythroid differentiation of primary human hematopoietic progenitor cells in vitro, even though it does not resemble erythropoietin, the natural ligand of this receptor. Here, we used a directed-evolution approach to explore the structural basis for the ability of transmembrane proteins to activate the hEPOR. A library that expresses thousands of mutants of the transmembrane activator was screened for variants that were more active than the original isolate at inducing growth factor independence in mouse cells expressing the hEPOR. The most active mutant, EBC5-16, supports erythroid differentiation in human cells with activity approaching that of EPO, as assessed by cell-surface expression of glycophorin A, a late-stage marker of erythroid differentiation. EBC5-16 contains a single isoleucine to serine substitution at position 25, which increases its ability to form dimers. Genetic studies confirmed the importance of dimerization for activity and identified the residues constituting the homodimer interface of EBC5-16. The interface requires a GxxxG dimer packing motif and a small amino acid at position 25 for maximal activity, implying that tight packing of the EBC5-16 dimer is a crucial determinant of activity. These experiments identified an artificial protein that causes robust activation of its target in a natural host cell, demonstrated the importance of dimerization of this protein for engagement of the hEPOR, and provided the framework for future structure-function studies of this novel mechanism of receptor activation.