The efficiency of reinitiation in mammalian translation systems depends in part on the size and arrangement of upstream open reading frames (upORFs). The gradual decrease in reinitiation as an upORF is lengthened, confirmed here using a variety of sequences, might reflect time-dependent loss of protein factors required for reinitiation. Consistent with the idea that the duration of elongation is what matters, reinitiation was nearly abolished when a pseudoknot that causes a pause in elongation was inserted into a short upORF. Control experiments showed that this transient pause in elongation had little effect on the final protein yield when the pseudoknot was moved from the upORF into the main ORF. Thus, the deleterious effect of slowing elongation is limited to the reinitiation mode. Another aspect of reinitiation investigated here is whether post-termination ribosomes can scan backwards to initiate at AUG codons positioned upstream from the terminator codon. Earlier studies that raised this possibility may have been complicated by the occurrence of leaky scanning along with reinitiation. Re-examination of the question, using constructs that preclude leaky scanning, shows barely detectable reinitiation from an AUG codon positioned 4 nt upstream from the terminator codon and no detectable reinitiation from an AUG codon positioned farther upstream. These experiments carried out with synthetic transcripts help to define the circumstances under which reinitiation may be expected to occur in the growing number of natural mRNAs that deviate from the simple first AUG rule.