The exon junction complex (EJC) is critical for mammalian nonsense-mediated mRNA decay and translational regulation. The EJC consists of numerous proteins, including four core proteins that are deposited on spliced mRNA during pre-mRNA splicing at a site that includes positions -20 to -24 of the 5' exon. However, the timing of deposition, how the deposition site is selected, and whether alternative sites of deposition occur are unknown. To address these questions, I created splicing substrates containing either DNA nucleotides or RNA secondary structure in the 5' exon. Using RNase H protection, toeprinting, and co-immunoprecipitation assays, I found that the site of EJC deposition shifts upstream when a stretch of DNA or RNA secondary structure appears at the canonical deposition site. These upstream shifts occur prior to exon ligation and are often accompanied by decreases in deposition efficiency, demonstrating that selection of an EJC deposition site is spatially constrained. These experiments also illustrate some of the requirements for EJC deposition, including the presence of ribose 2'OH groups. Although the EJC core protein eIF4AIII contacts four ribose 2'OH groups in crystal structures, I demonstrate that three 2'OH groups are necessary and sufficient for deposition. The requirement for three 2'OH groups is accompanied by a requirement for single-stranded nucleotides at the deposition site. Interestingly, if a stem-loop is present in the 5' exon at the normal site of deposition, then the EJC is efficiently deposited at least 25 nts upstream of the normal deposition site. However, if 2'H groups are present instead of 2'OH groups at the site of deposition, then the EJC is deposited with lower efficiency when exclusively deposited 12 or 16 nts upstream of the normal deposition site. Thus, the site of EJC deposition is more flexible than previously appreciated, and yet efficient deposition is spatially limited. I propose a model where the spatial limitation results from interactions between eIF4AIII and intron-associated proteins, which occur prior to and possibly during EJC assembly. These interactions would likely restrict the ability of eIF4AIII to make contacts with nucleotides not located close to the site of eIF4AIII's interactions with intron-associated proteins.