Antagonist antibodies targeting CD28 have been proposed as an alternative to the use of CD80/86 antagonists to modulate T cell responses in autoimmunity and transplantation. Advantages would be the blockade of CD28-mediated co-stimulatory signals without impeding the co-inhibitory signals dependent on CD80 interactions with CTLA-4 and PD-L1 that are important for the control of immune responses and for the function of regulatory T cells. Anti-CD28 antibodies are candidate antagonists only if they prevent access to the CD80/86 ligands without simultaneously stimulating CD28 itself, a process that is believed to depend on receptor multimerization. In this study, we evaluated the impact of different formats of a potentially antagonist anti-human CD28 antibody on T cell activation. In particular, we examined the role of valency and of the presence of an Fc domain, two components that might affect receptor multimerization either directly or in the presence of accessory cells expressing Fc receptors. Among monovalent (Fab', scFv), divalent (Fab'2), monovalent-Fc (Fv-Fc) and divalent-Fc (IgG) formats, only the monovalent formats showed consistent absence of induced CD28 multimerization and absence of associated activation of phosphoinositol-3-kinase, and clear antagonist properties in T cell stimulation assays. In contrast, divalent antibodies showed agonist properties that resulted in cell proliferation and cytokine release in an Fc-independent manner. Conjugation of monovalent antibodies with polyethylene glycol, α-1-antitrypsin or an Fc domain significantly extended their in vivo half-life without modifying their antagonist properties. In conclusion, these data indicate that monovalency is mandatory for maintaining the antagonistic activity of anti-CD28 monoclonal antibodies.